阅读说明：本技术 游戏道具的整理方法、装置和电子设备 (Game prop sorting method and device and electronic equipment ) 是由 冯潞潞 邹哲讷 尚悦 陶建容 范长杰 胡志鹏 于 2020-06-24 设计创作，主要内容包括：本发明提供了一种游戏道具的整理方法、装置和电子设备,响应于道具整理指令,获取道具整理指令对应的玩家的道具集合；从道具集合中获取至少一对道具对；针对每对道具对,将道具对中的第一道具的道具信息和第二道具的道具信息输入至预先训练完成的道具顺序模型,输出第一道具和第二道具的第一排序信息；根据每对道具对对应的第一排序信息,对道具集合中的道具进行排序。该方式中,通过道具顺序模型,自动学习背包中各个道具的顺序关系,基于该顺序关系对玩家背包中的道具进行排序,无需根据道具的编号进行排序,提高了背包中道具的排列顺序的灵活性,同时使得道具的排列顺序与玩家的个人习惯和游戏场景的实际需求相匹配,提高了玩家的游戏体验度。(The invention provides a method, a device and electronic equipment for arranging game props, which respond to a prop arrangement instruction and obtain a prop set of a player corresponding to the prop arrangement instruction; acquiring at least one pair of prop pairs from the prop set; for each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs. In the mode, the order relation of each prop in the backpack is automatically learned through the prop order model, the props in the player backpack are sequenced based on the order relation, sequencing is not needed according to the serial number of the props, the flexibility of the sequence of the props in the backpack is improved, meanwhile, the sequence of the props is matched with the personal habits of the players and the actual requirements of game scenes, and the game experience degree of the players is improved.)

1. A method for organizing play objects, the method comprising:

responding to a prop sorting instruction, and acquiring a prop set of a player corresponding to the prop sorting instruction;

obtaining at least one pair of prop pairs from the prop set;

for each pair of the props, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; wherein the item information is determined according to the current game scene and/or the current player;

and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of the props.

2. The method of claim 1, wherein the prop information is obtained by:

acquiring original information of a prop to be processed; the original information comprises a prop identification, a prop type, current game scene information and current player information of the prop to be processed;

and inputting the original information into an information encoder which is trained in advance to obtain the prop information of the prop to be processed.

3. The method of claim 2, wherein the information encoder is trained by:

obtaining a training sample containing a plurality of props;

inputting each prop in the training sample into an information encoder, and outputting initial prop information of each prop;

and inputting the initial prop information of each prop into an auxiliary training model so as to train the information encoder based on the auxiliary training model to obtain the trained information encoder.

4. The method of claim 3, wherein the training samples comprise: the method comprises the steps of obtaining a target prop, a comparison prop, first tag information and a use prop sequence in a sample prop set;

wherein, the target prop is a prop which has changed position;

the first tag information includes: the front-back position relation between the target prop and the comparison prop and whether the target prop and the comparison prop have an adjacent relation or not;

the use prop sequence is as follows: a sequence of props used by a player in the sample set of props since the position of the target prop has changed; the props in the prop using sequence are arranged according to the used time sequence.

5. The method of claim 4, wherein the auxiliary training model comprises a first auxiliary model and a second auxiliary model;

the step of inputting the initial prop information of each prop into an auxiliary training model to train the information encoder based on the auxiliary training model to obtain a trained information encoder includes:

inputting the initial prop information of each prop into a first auxiliary model, and outputting second sequencing information of the target props and the comparison props; training the information encoder and the first auxiliary model based on the first label information and the second sequencing information until the first auxiliary model converges to obtain an intermediate training result of the information encoder;

inputting each prop in the training sample into the intermediate training result, and outputting intermediate prop information of each prop;

inputting the middle prop information of each prop into a second auxiliary model, and outputting position information indicating whether the target prop and the comparison prop are adjacent or not; and training the information encoder and a second auxiliary model based on the first label information and the position information until the second auxiliary model converges to obtain the trained information encoder.

6. The method of claim 5, wherein the first auxiliary model comprises: a first fully connected network, a second fully connected network, a first sequence network, and a third fully connected network;

the step of inputting the initial prop information of each prop into a first auxiliary model and outputting the target props and the second sequencing information of the comparison props comprises:

inputting initial prop information of the target prop into the first fully-connected network, and outputting first intermediate information; inputting the initial prop information of the comparison prop into the second fully-connected network, and outputting second intermediate information;

inputting the initial prop information of the prop using sequence into the first sequence network, and outputting third intermediate information;

inputting the first intermediate information, the second intermediate information and the third intermediate information into the third fully-connected network to obtain a first output result, and determining second sequencing information of the target prop and the comparison prop based on the first output result; the second ranking information includes: a probability that the target prop is arranged in front of the comparison prop.

7. The method of claim 5, wherein the second auxiliary model comprises: a fourth fully connected network, a fifth fully connected network, a second sequence network, and a sixth fully connected network;

the step of inputting the intermediate prop information of each prop into a second auxiliary model and outputting position information indicating whether the target prop and the comparison prop are adjacent to each other includes:

inputting the intermediate prop information of the target prop into the fourth fully-connected network, and outputting fourth intermediate information; inputting the intermediate prop information of the comparison prop into the fifth fully-connected network, and outputting fifth intermediate information;

inputting the intermediate prop information of the prop using sequence into the second sequence network, and outputting sixth intermediate information;

inputting the fourth intermediate information, the fifth intermediate information and the sixth intermediate information into the sixth fully-connected network to obtain a second output result, and determining position information indicating whether the target prop and the comparison prop are adjacent or not based on the second output result; the location information includes: and the probability that the target prop is adjacent to the comparison prop.

8. The method of claim 1, wherein the prop order model comprises a seventh fully connected network, an eighth fully connected network, and a ninth fully connected network;

inputting the prop information of a first prop and the prop information of a second prop in the prop pair into a pre-trained prop sequence model, and outputting the first prop and the first sequence information of the second prop, wherein the steps comprise:

inputting prop information of the first prop to the seventh fully-connected network, and outputting seventh intermediate information; inputting prop information of the second prop to the eighth fully-connected network, and outputting eighth intermediate information;

inputting the seventh intermediate information and the eighth intermediate information to the ninth fully-connected network to obtain a third output result; determining first ranking information for the first prop and the second prop based on the third output result; wherein the first ordering information includes: a probability that the first prop is arranged in front of the second prop, or a ranking of the first prop and the second prop.

9. The method of claim 8, wherein the prop sequence model is trained by:

obtaining a sample prop pair and second label information; the sample prop pair comprises a target prop and a comparison prop in a sample prop set; wherein, the target prop is a prop which has changed position; the second tag information includes: the front-back position relation between the target prop and the comparison prop;

inputting prop information of the target prop into the seventh fully-connected network, and outputting seventh intermediate information; inputting the prop information of the comparison prop into the eighth fully-connected network, and outputting eighth intermediate information;

inputting the seventh intermediate information and the eighth intermediate information into the ninth fully-connected network to obtain a third output result, and obtaining third ordering information of the target prop and the comparison prop based on the third output result;

training the prop sequence model based on the second label information and the third sequencing information until the prop sequence model converges to obtain the trained prop training model.

10. The method of claim 1, wherein said step of obtaining at least one pair of items from said set of items comprises:

clustering the props in the prop set to obtain at least one prop cluster; wherein each prop type cluster comprises at least one prop in the prop set; aiming at each prop type cluster, if a plurality of props are included in the current prop type cluster, generating at least one pair of prop pairs based on the plurality of props;

the step of sequencing the props in the prop set according to the first sequencing information corresponding to each pair of the prop pairs comprises:

counting the times of the first arrangement of each prop in each prop pair in the current prop cluster according to the first arrangement information corresponding to each pair of prop pairs in the current prop cluster; and sequencing the props in the current prop type cluster based on the number of times that each prop is arranged in front of each prop pair.

11. The method of claim 10, wherein after the step of ranking the items in the current item cluster based on the number of times each item ranks first in each item pair, the method further comprises:

setting current scene information in the prop information of each prop in the current prop cluster to be zero, and obtaining temporary prop information of each prop;

for each pair of the prop pairs in the current prop cluster, inputting temporary prop information of a first prop and temporary prop information of a second prop in the prop pairs into the prop sequence model, and outputting temporary sequencing information of the first prop and the second prop;

sequencing the props in the prop set according to the temporary sequencing information corresponding to each pair of the props to obtain a temporary sequencing result;

and determining the props related to the current scene in the current props cluster based on the temporary sequencing result.

12. The method of claim 10, wherein after the step of ranking the items in the current item cluster based on the number of times each item ranks first in each item pair, the method further comprises:

determining the props which are arranged in front of all prop pairs in the current prop cluster most frequently as representative props in the current prop cluster aiming at each prop cluster;

generating at least one pair of representative prop pairs based on the representative props of each prop class cluster;

counting the times of each representative prop being arranged in front of each representative prop pair according to the first sequencing information corresponding to each representative prop pair;

and sequencing each representative prop cluster based on the number of times each representative prop is arranged in front of each representative prop pair.

13. The method of claim 1, wherein after the step of sorting the props in the set of props according to the first sorting information corresponding to each pair of the props, the method further comprises:

responding to a prop adjusting instruction, and acquiring a prop to be adjusted and a target position corresponding to the prop adjusting instruction; and adjusting the prop to be adjusted in the prop set to the target position.

14. An organizing device for play objects, the device comprising:

the system comprises a prop set acquisition module, a display module and a display module, wherein the prop set acquisition module is used for responding to a prop arrangement instruction and acquiring a prop set of a player corresponding to the prop arrangement instruction;

the prop pair obtaining module is used for obtaining at least one pair of prop pairs from the prop set;

the sequencing information output model is used for inputting the prop information of a first prop and the prop information of a second prop in each pair of prop pairs into a pre-trained prop sequence model and outputting the first sequencing information of the first prop and the second prop; wherein the item information is determined according to the current game scene and/or the current player;

and the prop sequencing module is used for sequencing the props in the prop set according to the first sequencing information corresponding to each pair of the props.

15. An electronic device comprising a processor and a memory, the memory storing machine executable instructions executable by the processor, the processor executing the machine executable instructions to implement the method of organizing play objects of any of claims 1-13.

16. A machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to implement the method of organizing play objects of any of claims 1-13.

Technical Field

The invention relates to the technical field of game props, in particular to a method and a device for arranging game props and electronic equipment.

Background

The game world has abundant props, and game players can continuously obtain various props in the game experience process; the props acquired by the players are placed in props backpacks of the players. Because the space of the prop backpack is limited, the props in the backpack need to be often arranged. In the related art, a game plan sets a prop number for each item, and in the process of arranging props, the props in the backpack are sorted according to the sequence of the prop numbers. However, when the props in the game are continuously increased, the task of setting numbers for the props becomes more complicated and error-prone, and if the numbers are set incorrectly, the arrangement of the props in the process of arranging the props is disordered; in addition, the mode of arranging the props based on the prop numbers ensures that the arrangement sequence of the props in the backpack is relatively solidified and is difficult to match with the personal habits of the players and the actual requirements of the game scenes, and the game experience of the players is reduced.

Disclosure of Invention

In view of this, the present invention provides a method, an apparatus and an electronic device for organizing game items, so as to improve the game experience of a user.

In a first aspect, an embodiment of the present invention provides a method for organizing game props, where the method includes: responding to the item arrangement instruction, and acquiring an item set of the player corresponding to the item arrangement instruction; acquiring at least one pair of prop pairs from the prop set; for each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; wherein, the prop information is determined according to the current game scene and/or the current player; and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs.

Further, the prop information is obtained by the following method: acquiring original information of a prop to be processed; the original information comprises a prop identification, a prop type, current game scene information and current player information of the prop to be processed; and inputting the original information into an information encoder which is trained in advance to obtain the prop information of the prop to be processed.

Further, the information encoder is trained by: obtaining a training sample containing a plurality of props; inputting each prop in the training sample into an information encoder, and outputting initial prop information of each prop; and inputting the initial prop information of each prop into an auxiliary training model so as to train the information encoder based on the auxiliary training model to obtain the trained information encoder.

Further, the training samples include: the method comprises the steps of obtaining a target prop, a comparison prop, first tag information and a use prop sequence in a sample prop set; wherein, the target prop is a prop which has changed position; the first tag information includes: the front-back position relation between the target prop and the comparison prop and whether the target prop and the comparison prop have an adjacent relation; the using prop sequence is as follows: the method comprises the steps that after a position change occurs to a target prop, a prop sequence used by a player in a sample prop set is obtained; the props in the prop use sequence are arranged according to the used time sequence.

Further, the auxiliary training model comprises a first auxiliary model and a second auxiliary model; inputting the initial prop information of each prop into an auxiliary training model so as to train an information encoder based on the auxiliary training model, and obtaining the trained information encoder, wherein the method comprises the following steps: inputting the initial prop information of each prop into a first auxiliary model, and outputting a target prop and second sequencing information in front of the comparison prop; training the information encoder and the first auxiliary model based on the first label information and the second sequencing information until the first auxiliary model converges to obtain an intermediate training result of the information encoder; inputting each prop in the training sample into an intermediate training result, and outputting intermediate prop information of each prop; inputting the middle prop information of each prop into a second auxiliary model, and outputting position information indicating whether the target prop and the comparison prop are adjacent or not; and training the information encoder and the second auxiliary model based on the first label information and the position information until the second auxiliary model converges to obtain the trained information encoder.

Further, the first auxiliary model includes: a first fully connected network, a second fully connected network, a first sequence network, and a third fully connected network; inputting the initial prop information of each prop into a first auxiliary model, and outputting the target props and comparing the second sequencing information of the props, wherein the method comprises the following steps of: inputting initial prop information of a target prop into a first fully-connected network, and outputting first intermediate information; inputting the initial prop information of the comparison props into a second fully-connected network, and outputting second intermediate information; inputting initial prop information using a prop sequence into the first sequence network, and outputting third intermediate information; inputting the first intermediate information, the second intermediate information and the third intermediate information into a third fully-connected network to obtain a first output result, and determining a target prop and comparing second sequencing information of the props based on the first output result; the second ranking information includes: the probability that the target prop is arranged in front of the comparison prop.

Further, the second auxiliary model includes: a fourth fully connected network, a fifth fully connected network, a second sequence network, and a sixth fully connected network; inputting the middle prop information of each prop into a second auxiliary model, and outputting position information indicating whether the target prop and the comparison prop are adjacent, wherein the step comprises the following steps of: inputting the intermediate prop information of the target prop into a fourth fully-connected network, and outputting fourth intermediate information; inputting the intermediate prop information of the comparison props into a fifth fully-connected network, and outputting fifth intermediate information; inputting the intermediate prop information using the prop sequence into a second sequence network, and outputting sixth intermediate information; inputting the fourth intermediate information, the fifth intermediate information and the sixth intermediate information into a sixth fully-connected network to obtain a second output result, and determining position information indicating whether the target prop and the comparison prop are adjacent or not based on the second output result; the location information includes: and the probability that the target prop is adjacent to the comparison prop.

Further, the prop sequence model comprises a seventh fully connected network, an eighth fully connected network and a ninth fully connected network; inputting the prop information of the first prop and the prop information of the second prop in the prop pair into a prop sequence model which is trained and completed in advance, and outputting the first sequence information of the first prop and the second prop, wherein the method comprises the following steps: inputting the prop information of the first tool to a seventh fully-connected network, and outputting seventh intermediate information; inputting prop information of the second prop to an eighth fully-connected network, and outputting eighth intermediate information; inputting the seventh intermediate information and the eighth intermediate information to a ninth fully-connected network to obtain a third output result; determining first sequencing information of the first prop and the second prop based on the third output result; wherein the first ordering information includes: the first prop has a probability of being arranged in front of the second prop, or the result of the ordering of the first prop and the second prop.

Further, the prop sequence model is obtained by training in the following way: obtaining a sample prop pair and second label information; the sample prop pair comprises a target prop and a comparison prop in the sample prop set; wherein, the target prop is a prop which has changed position; the second tag information includes: the front-back position relation between the target prop and the comparison prop; inputting prop information of the target prop into a seventh fully-connected network, and outputting seventh intermediate information; inputting the prop information of the comparison props into an eighth fully-connected network, and outputting eighth intermediate information; inputting the seventh intermediate information and the eighth intermediate information into a ninth fully-connected network to obtain a third output result, and obtaining third sequencing information of the target prop and the comparison prop based on the third output result; and training the prop sequence model based on the second label information and the third sequencing information until the prop sequence model converges to obtain a trained prop training model.

Further, the step of obtaining at least one pair of prop pairs from the prop set includes: clustering the props in the prop set to obtain at least one prop cluster; each prop type cluster comprises at least one prop in a prop set; aiming at each prop type cluster, if the current prop type cluster comprises a plurality of props, generating at least one pair of prop pairs based on the plurality of props; according to the first sequencing information corresponding to each pair of prop pairs, sequencing the props in the prop set, and the method comprises the following steps: counting the times of each prop in the current prop cluster being arranged in front of each prop pair according to the first sequencing information corresponding to each pair of prop in the current prop cluster; and sequencing the props in the current prop type cluster based on the number of times that each prop is arranged in front of each prop pair.

Further, after the step of sorting the props in the current prop type cluster based on the number of times each prop is ranked in front of each prop pair, the method further comprises: setting current scene information in the prop information of each prop in the current prop cluster to be zero, and obtaining temporary prop information of each prop; aiming at each pair of prop pairs in the current prop cluster, inputting the temporary prop information of a first prop and the temporary prop information of a second prop in the prop pairs into a prop sequence model, and outputting the temporary sequencing information of the first prop and the second prop; sequencing the props in the prop set according to the temporary sequencing information corresponding to each pair of props to obtain a temporary sequencing result; and determining the props related to the current scene in the current props cluster based on the temporary sequencing result.

Further, after the step of sorting the props in the current prop type cluster based on the number of times each prop is ranked in front of each prop pair, the method further comprises: determining the props which are arranged in front of all prop pairs in the current prop cluster most frequently as representative props in the current prop cluster aiming at each prop cluster; generating at least one pair of representative prop pairs based on the representative props of each prop type cluster; counting the times of each representative prop being arranged in front of each representative prop pair according to the first sequencing information corresponding to each representative prop pair; each prop category cluster is ranked based on the number of times each representative prop is ranked first in each representative prop pair.

Further, after the step of sorting the props in the prop set according to the first sorting information corresponding to each pair of prop pairs, the method further includes: responding to the prop adjusting instruction, and acquiring a prop to be adjusted and a target position corresponding to the prop adjusting instruction; and adjusting the prop to be adjusted in the prop set to a target position.

In a second aspect, an embodiment of the present invention provides an arrangement device for a game item, where the device includes: the property collection obtaining module is used for responding to the property arrangement instruction and obtaining a property collection of the player corresponding to the property arrangement instruction; the prop pair obtaining module is used for obtaining at least one pair of prop pairs from the prop set; the sequencing information output model is used for inputting the prop information of a first prop and the prop information of a second prop in each pair of prop pairs into a pre-trained prop sequence model and outputting first sequencing information of the first prop and the second prop; wherein, the prop information is determined according to the current game scene and/or the current player; and the prop sequencing module is used for sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs.

In a third aspect, an embodiment of the present invention provides an electronic device, which includes a processor and a memory, where the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the method for organizing game items described above.

In a fourth aspect, embodiments of the present invention provide a machine-readable storage medium storing machine-executable instructions, which when invoked and executed by a processor, cause the processor to implement the method for organizing play objects described above.

The embodiment of the invention has the following beneficial effects:

the embodiment of the invention provides a game item arrangement method, a game item arrangement device and electronic equipment, wherein a player item set corresponding to an item arrangement instruction is obtained in response to the item arrangement instruction; then at least one pair of prop pairs is obtained from the prop set; for each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs. In the mode, the order relation of each prop in the backpack is automatically learned through the prop order model, the props in the player backpack are sequenced based on the order relation, sequencing is not needed according to the serial number of the props, the flexibility of the sequence of the props in the backpack is improved, meanwhile, the sequence of the props is matched with the personal habits of the players and the actual requirements of game scenes, and the game experience degree of the players is improved.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a flow chart of a method for organizing play objects according to an embodiment of the present invention;

FIG. 2 is a flowchart of a training method for an information encoder according to an embodiment of the present invention;

FIG. 3 is a schematic diagram illustrating movement of props in a sample prop set according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a complete training sample according to an embodiment of the present invention;

FIG. 5 is a flowchart illustrating a training method of an information encoder according to an embodiment of the present invention;

FIG. 6 is a flow chart of another training method for an information encoder according to an embodiment of the present invention;

FIG. 7 is a flowchart illustrating a training method of a first auxiliary model according to an embodiment of the present invention;

FIG. 8 is a flowchart illustrating a training method for a second auxiliary model according to an embodiment of the present invention;

fig. 9 is a schematic flowchart of a training method of a road furniture sequence model according to an embodiment of the present invention;

fig. 10 is a schematic diagram of a clustering result according to an embodiment of the present invention;

FIG. 11 is a timing diagram illustrating a specific method for organizing play objects according to an embodiment of the present invention;

FIG. 12 is a schematic structural diagram of an organizing device for play objects according to an embodiment of the present invention;

fig. 13 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

To make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

At present, in a traditional method for arranging props in a backpack, numbers are usually configured for each prop in advance by planning, and when a user has a demand for arranging the props by one key, a system sorts all the props of the user according to the numbers of the props and displays the props to the user; when new props are continuously added in the game, the planning needs to manually configure numbers for the new road tools, and when the number of the road tools is large, the planning is difficult to make proper configuration; in addition, different players have different habits on the storage modes of the property classification, the personalized requirements of part of users cannot be met by using the property numbers in a fixed sequence, and the game experience of the players is low. Based on this, the method, the device and the electronic device for arranging the game props provided by the embodiment of the invention can be applied to various game scenes, and particularly can be applied to game scenes for arranging the game props.

In order to facilitate understanding of the embodiment, a detailed description is first given of a method for organizing game props disclosed in the embodiment of the present invention, where the method is applicable to a terminal device, a graphical user interface is obtained by executing a software application on a processor of the terminal device and rendering on a display of the terminal device, and the graphical user interface is provided with a skill control; wherein the software application may be a gaming application; the graphical user interface may be a game scene currently being presented in a game; the skill control in the image user interface can be used for arranging props in the backpack in a one-key mode.

As shown in fig. 1, the method comprises the steps of:

step S102, responding to the item arrangement instruction, and acquiring an item set of the player corresponding to the item arrangement instruction;

the prop sorting instruction can be triggered by a user through a prop sorting control; the prop finishing control is usually arranged at a relatively fixed position in a game scene; in equipment controlled by a touch screen, such as a mobile phone and a tablet personal computer, a game player can click a prop finishing control by a finger; in terminal equipment such as a desktop computer, a notebook computer and the like, a game player can click a prop finishing control through a mouse, an external touch screen and the like; in the device controlled by voice, the game player can also control the trigger of the prop finishing control through voice input devices such as a microphone and the like. The item set generally refers to information of all items in a backpack of a game player, and the item set may include information such as item sequences, current game scenes, user images, and the like; the method can also comprise information such as prop identification, prop types, current game scenes, user pictures and the like; the method can also comprise a property identification code, a property type code, a current game scene code, a user portrait code and the like; the user representation comprises information such as gender, age, online duration, consumption level and the like of the user.

Specifically, after responding to the prop arrangement instruction, the client may send the prop sequence, the current scene, the user image and other information in the player backpack corresponding to the prop arrangement instruction to the server, and may perform encoding processing on the props in the player backpack by using data processing and other modes to obtain the prop set.

Step S104, obtaining at least one pair of prop pairs from the prop set;

the prop set usually comprises a plurality of props; the prop pair can be two props with similar functions, can also be two props with the same type, can also be two props with completely different functions and the like; the prop pair can be randomly extracted from a prop set, and two props of the same or different types can also be selected in a clustering mode. If a plurality of pairs of prop pairs are obtained, repeated props can be arranged between each pair of prop pairs, or no repeated props can be arranged between the pairs of prop pairs.

Step S106, aiming at each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; wherein, the prop information is determined according to the current game scene and/or the current player;

the property information generally comprises properties, property types, property using scenes, user figures and other information; the current game scenes include various scenes, such as a scene in which a little monster is combated in the copy, a scene in which a big monster is combated in the copy, a scene in which a helmet is being forged, a scene in which clothes is being forged, and the like; in addition, the games in the copy are the same, some games need to distinguish scenes of different enemy games, and some games do not distinguish; meanwhile, some games need to distinguish different copies, and some games consider all the copies to belong to the same scene.

In order to enable the pre-trained prop sequence model to understand the input data, the prop information generally includes a prop code, a prop type code, a scene code using the prop, and a user code; the above-mentioned prop sequence model trained in advance may be various Network models, such as CNN (Convolutional Neural Networks), RNN (Recurrent Neural Networks), DNN (Deep Neural Networks), and the like; the prop sequence model can judge the sequence of the input first prop and the input second prop according to the trained network parameters, and the output first sequencing information of the first prop and the second prop can be the probability that the first prop is arranged in front relative to the second prop, and the sequence of the first prop and the second prop in the backpack can be determined according to the probability; or may be a specific arrangement result of the first prop and the second prop, for example, the specific arrangement result may be: the first item is arranged in front of the second item, or the second item is arranged in front of the first item.

Since each item can be used in various game scenarios, the item information of each item is different in different game scenarios and different game players. Therefore, the item information of each item can be determined according to the current game scene and the current player, or according to the current game scene, or according to the current player; for example, the item information of item a in scene b where the current player is k generally includes information of current player being k and information of game scene b, and the like, for example, the item information may be information of item a relative to player k in scene b; the information may be various information encodings, etc.

And S108, sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs.

Usually, the prop set includes a plurality of props, the arrangement order of each pair of props can be determined according to the first arrangement information corresponding to each pair of props, when each pair of props has a repeated prop, the repeated props can be utilized to arrange the sequence between each pair of props, and finally all props in the prop set are arranged. If each pair of prop pairs does not have repeated props, the props between the prop pairs can be sorted according to the type of each prop in the prop pairs, and finally all the props in the prop set are sorted.

The embodiment of the invention provides a method for arranging game props, which is used for responding to a prop arrangement instruction and acquiring a prop set of a player corresponding to the prop arrangement instruction; then at least one pair of prop pairs is obtained from the prop set; for each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs. In the mode, the order relation of each prop in the backpack is automatically learned through the prop order model, the props in the player backpack are sequenced based on the order relation, sequencing is not needed according to the serial number of the props, the flexibility of the sequence of the props in the backpack is improved, meanwhile, the sequence of the props is matched with the personal habits of the players and the actual requirements of game scenes, and the game experience degree of the players is improved.

The above embodiments describe that the item information may be determined according to the current game scenario and the current player, or may be determined based on either of the two; in the method, the item information not only needs to refer to the current game scene and the relevant information of the current player, but also needs to refer to the relevant information of the item, and the specific item information obtaining method is realized by the following method:

step A1, acquiring original information of the prop to be processed; the original information comprises a prop identification, a prop type, current game scene information and current player information of the prop to be processed;

the prop to be processed can be each prop in a current player game backpack; the item identifiers are usually ID (identification number) data without numerical significance, for example, "3001", "3002" and "3003" represent three item identifiers in the game respectively; the item type and the information of the current game scene are the same as the item identification type, and are also ID data, for example, "a 001" and "a 002" represent two item types in the game respectively; for another example, "P001" and "P002" represent two different game scenes in the game, respectively; the current player information comprises information such as gender, age, online duration, consumption level and the like; the current player information may be numerical data, for example, the player gender may be "00" and "01", wherein "00" represents a female and "01" represents a male; the age of the player can be directly expressed according to the age value; the online time length can also be directly expressed according to a time value; the consumption level may be expressed in terms of a specific value of consumption.

For example, the player clicks the prop sorting control, and the client responds to the prop sorting instruction, and obtains the original information of the prop to be processed from the log data.

And step A2, inputting the original information into the pre-trained information encoder to obtain the property information of the property to be processed.

Specifically, for original information of a prop to be processed, which is obtained from log data of a client, a prop sequence model may not be directly identified, and the original information needs to be preprocessed; the original information can be input into an information encoder which is trained in advance through equipment such as a server, and the prop identification, the prop type, the current game scene information and the current player information in the original information are numbered through the information encoder; for example, the ID data such as the item identifier, the item type, and the current game scenario information may be sorted according to the dictionary sequence, and numbered according to 1, 2, 3, and … N; the user image information is generally numerical data, and no special processing is required for numbering.

In the method, a method for generating prop information in an online service module is provided, the online service module is applied by an information coding model which is trained in advance, original information is input into an information coder which is trained in advance, prop information of a prop to be processed is obtained, and therefore the model for subsequent prop arrangement carries out prop arrangement operation.

The embodiment also provides another method for organizing game props, and a specific implementation manner of the training of the information encoder is described in the method. Fig. 2 shows a flow chart of a training method of an information encoder, which includes the following steps:

step S202, obtaining a training sample containing a plurality of props;

the training samples of the plurality of props can be user data of all users in the current game collected in an offline state; specifically, in an off-line state, training samples are obtained from a log database, wherein the training samples comprise information such as a property using sequence of a player, user portrait information, player mobile property data, player backpack data and the like; the use item sequence represents a sequence formed by arranging items used by a player in a time sequence from a certain moment; for example, the sequence starting from time t is called the prop-using sequence at time t; the user portrait information comprises gender, age, online duration, consumption level and the like; the player moves the prop data, including the moved prop, the position before moving, the position after moving and the scene during moving; the player backpack data includes the location of each prop in the backpack.

In a preferred embodiment, the training sample includes: the method comprises the steps of obtaining a target prop, a comparison prop, first tag information and a use prop sequence in a sample prop set; wherein, the target prop is a prop which has changed position; the first tag information includes: the front-back position relation between the target prop and the comparison prop and whether the target prop and the comparison prop have an adjacent relation; the using prop sequence is as follows: the method comprises the steps that after a position change occurs to a target prop, a prop sequence used by a player in a sample prop set is obtained; the props in the prop use sequence are arranged according to the used time sequence.

Usually, the server records the data of each item moved by the user, and a set of training data can be constructed according to the record of each item moved. Wherein moving a prop can generally be divided into two situations: firstly, moving a prop to an empty position; secondly, changing the position of the prop and another prop; the record of the prop exchange position can be processed into two pieces of training data, namely, the two props are respectively moved to an empty position; therefore, the mobile prop is the target prop.

The comparison prop can be any prop in the user backpack except the target prop; the first tag information may be represented by "0", "1", "2", and "3", for example, the target prop is at a front position of the comparison prop, at this time, the first tag information may be "1", the target prop is at a rear position of the comparison prop, at this time, the first tag information may be "0", the target prop is adjacent to the comparison prop, at this time, the first tag information may be "3", the target prop is not adjacent to the comparison prop, at this time, the first tag information may be "2".

The position of each prop in the backpack of the user can be in a matrix arrangement mode, and the adjacent relation can comprise that two props are within three positions away from each other in the backpack; the distance between the three positions may be manhattan distance (manhattan distance), that is, the transverse distance plus the longitudinal distance is less than or equal to three, and at this time, two props may be considered to be adjacent.

For example, refer to the schematic diagram of the movement of a prop in a sample prop set shown in fig. 3; in the sample set, the prop a is moved to an empty position, and the prop b and the prop c are exchanged in position, so that a movable prop sequence comprising the prop a, the prop b and the prop c is formed; after the position of the used prop sequences of prop a, prop b and prop c is changed, the prop sequences used by the player in the sample prop set comprise prop d, prop e, prop f and prop g. In this embodiment, after the prop a is moved by the user X, a training sample is constructed as an example; firstly, a movable prop a is a prop with a changed position, namely a target prop; after the prop a is moved, the user only uses the prop d, the prop e, the prop f and the prop g, so that the prop d, the prop e, the prop f and the prop g are determined as a prop using sequence; the comparison prop is any prop except the prop a, and can be any prop of a prop d, a prop e, a prop f, a prop b, a prop g and a prop c; the first tag information may be "1" or "0" for the front-back position relationship between the target prop and the comparison prop through the relationship between the comparison target prop and the comparison prop; the first tag information may be "2" or "3" as to whether the target prop and the comparison prop have an adjacent relationship.

So after user X removed stage property a, to the front and back position relation of target stage property and comparison stage property, can constitute 6 training samples:

the target prop: a, comparing props: d, using prop sequence: d. e, f, g, label: 0;

the target prop: a, comparing props: e, using a prop sequence: d. e, f, g, label: 0;

the target prop: a, comparing props: f, using a prop sequence: d. e, f, g, label: 1;

the target prop: a, comparing props: b, using a prop sequence: d. e, f, g, label: 1;

the target prop: a, comparing props: g, using a prop sequence: d. e, f, g, label: 1;

the target prop: a, comparing props: c, using a prop sequence: d. e, f, g, label: 1;

similarly, after the user X exchanges the props b and c, the user X can be split into a target prop b and a target prop c to respectively generate 6 training data; wherein, user X removes stage property b after, to target stage property and the front and back position relation of comparing the stage property, can constitute 6 training samples:

the target prop: b, comparing the props: a, using a prop sequence: f. g, labeling: 0;

the target prop: b, comparing the props: d, using prop sequence: f. g, labeling: 0;

the target prop: b, comparing the props: e, using a prop sequence: f. g, labeling: 0;

the target prop: b, comparing the props: f, using a prop sequence: f. g, labeling: 1;

the target prop: b, comparing the props: g, using a prop sequence: f. g, labeling: 1;

the target prop: b, comparing the props: c, using a prop sequence: f. g, labeling: 1;

after user X moves prop c, 6 training samples can be formed for the front-back position relation of the target prop and the comparison prop:

the target prop: c, comparing the props: a, using a prop sequence: f. g, labeling: 0;

the target prop: c, comparing the props: b, using a prop sequence: f. g, labeling: 1;

the target prop: c, comparing the props: d, using prop sequence: f. g, labeling: 0;

the target prop: c, comparing the props: e, using a prop sequence: f. g, labeling: 0;

the target prop: c, comparing the props: f, using a prop sequence: f. g, labeling: 0;

the target prop: c, comparing the props: g, using a prop sequence: f. g, labeling: 1;

a total of 18 training samples are generated in the sample prop set of the end user X; in addition, when a training sample is constructed, three additional prop information needs to be added to each prop to form a prop quadruple which is respectively a type of the prop, a scene of the prop and a user image; for the target prop and the comparison prop, the scene is the scene when the prop is moved, and for the use prop, the scene is the scene when the prop is used; to "target prop: a, comparing props: d, using prop sequence: d. e, f, g, label: for example, referring to fig. 4, a structural diagram of a complete training sample is shown, where the training sample includes quadruple information of each prop and also includes first tag information.

The pseudo code that constructs the training samples can be expressed as:

for all users:

obtaining a user representation

All the mobile prop records of For:

obtaining a sequence of using props

Target property (moved property)

For all non-moving props:

obtain comparison prop (other props)

Obtaining the first label information (judging the position relation of the target prop and the comparison prop)

A piece of training data is constructed.

In addition, in the use stage of the information coding model, aiming at a certain user, a prop adjustment record of the user is continuously collected, and the information coding model continues to be trained on the basis of the prop adjustment record, so that the information coding model can learn the prop storage habit of the user; the properties are personalized and sorted aiming at different users, so that the personalized requirements of different players are met.

Step S204, inputting each prop in the training sample into an information encoder, and outputting initial prop information of each prop;

the initial prop information of each prop may be represented as a code of the prop relative to the user in the scene, and referring to a flow diagram of a training mode of an information encoder shown in fig. 5, the prop, the type of the prop, the scene, and the user image are encoded by using a prop encoder, a prop type encoder, a scene encoder, and a user encoder, respectively; and splicing the coded prop codes, the channel a type codes, the scene codes and the user codes, and processing by using a full-connection network to obtain the codes of the props relative to the users in the scene. The property encoder, the property type encoder and the scene encoder in the information encoder can be realized by an Embedding network, and the user encoder can be realized by a full-connection network; wherein, the Embedding network indicates to input an ID data and output a vector representation after ID coding; the Embedding network needs to be trained to get meaningful vectors.

Taking the 18 training samples formed in the above steps as an example, player X uses property a in scene a; and (3) the' target prop: a, comparing props: d, using prop sequence: d. e, f, g, label: each prop in the training sample of 0 ″ is input into the information encoder, and the initial prop information of each prop is output as the code of the prop a in the scene a relative to the user X, and the code can be data such as numerical values, letters and the like.

Step S206, inputting the initial prop information of each prop into an auxiliary training model, and training an information encoder based on the auxiliary training model to obtain the trained information encoder.

The number of the auxiliary training models can be one or more, and different auxiliary training models have different prop training functions; the initial prop information of each prop can be input into an auxiliary training model, and parameters in the information encoder are trained by using the auxiliary training model and the output result of the information encoder. For example, the information encoder may be trained using a back propagation algorithm, a forward propagation algorithm, etc., so that the trained information encoder has different information encoding modes. In addition, training the information coding model can be understood as training the props, the prop types, the scenes, the four encoders corresponding to the users, and the parameters of the 5 parts in the first fully-connected network.

In the method, the data of the props moved by the users each time are recorded, and training samples of the props of all the users in the game can be constructed according to the recorded data, wherein the training samples comprise target props, comparison props, first label information and information of prop sequence use; inputting the prop information of each prop into an information coding model and an auxiliary model, training the information coding model based on the auxiliary model, so that the trained information coding model can learn the optimal coding mode; the flexibility of the arrangement sequence of the props in the backpack is improved, so that the personal habits of the players are matched with the actual requirements of the game scenes, and the game experience of the players is improved.

The embodiment also provides another game item sorting method, and the method describes a specific implementation manner of inputting the initial item information of each item into an auxiliary training model so as to train an information encoder based on the auxiliary training model to obtain the trained information encoder. The auxiliary training model comprises a first auxiliary model and a second auxiliary model.

As shown in fig. 6, this method includes the steps of:

step S602, inputting the initial prop information of each prop into a first auxiliary model, and outputting a target prop and second sequencing information of the comparison props; training the information encoder and the first auxiliary model based on the first label information and the second sequencing information until the first auxiliary model converges to obtain an intermediate training result of the information encoder;

the first auxiliary model may be various Network models, such as CNN (Convolutional Neural networks), RNN (Recurrent Neural networks), DNN (deep Neural networks), and the like; the first auxiliary model is typically a model that has not been trained; in addition, when the order of the props at a certain moment is predicted, the data of the props used after the moment are utilized, and the first auxiliary model can also be called a posterior prop order model; therefore, the first tag information in the initial prop information of each prop input to the first auxiliary model is the front-back position relationship between the target prop and the comparison prop, and can be represented by "0" and "1". The second sorting information may be a probability that the target property is arranged in front of the comparison property, or a result of arrangement of the target property and the comparison property, for example, the target property is arranged in front of the comparison property, or the comparison property is arranged in front of the target property.

Specifically, the first label information is compared with second arrangement information output by the first auxiliary model, and learnable parameters in the first auxiliary model and the information coding model are trained by using a back propagation technique and the like, wherein the learnable parameters include: scientific system parameters in the information coding model and the first auxiliary model are determined until the first auxiliary model converges, so that the first auxiliary model can sequence the precedence relationship of the props in the backpack; after the first auxiliary model converges, an intermediate training result of the information encoder can be obtained, and the intermediate training result can be understood that the information encoding model can learn an optimal encoding mode from the data of the sequence of the props while the first auxiliary model is trained; through the coding mode, props at different positions in the backpack can be distinguished; and obtaining the intermediate training result of the information coding model.

In addition, the first auxiliary model convergence may be that the first auxiliary model has no significant difference in performance before and after training, i.e., the first auxiliary model is considered to have converged; the first auxiliary model can also be set with a loss value, and when the loss value output for each input data is fixed or very close to a certain fixed value, the model is considered to be converged; the condition that the loss value satisfies a preset loss threshold may be further used as a condition for convergence of the first auxiliary model.

In a preferred embodiment, the first auxiliary model includes: a first fully connected network, a second fully connected network, a first sequence network, and a third fully connected network; aiming at the steps of inputting the initial prop information of each prop into the first auxiliary model, outputting the target props and comparing the second sequencing information of the props, the method comprises the following steps:

step B1, inputting the initial prop information of the target prop into the first fully-connected network, and outputting first intermediate information; inputting the initial prop information of the comparison props into a second fully-connected network, and outputting second intermediate information;

referring to fig. 7, a flow chart of a training method of a first auxiliary model is shown, wherein parameters in an encoder in a first fully-connected network, a second fully-connected network, and an information coding model need to be learned; the information coding model in the figure is the information coding model being trained. In actual implementation, the information coding model and the first auxiliary model may be spliced, all the prop quadruples in the input data are coded by the information coding model, and then the output result of the information coding model, that is, the initial prop information, is used as the input of the first auxiliary model, which includes target prop coding, comparison prop coding, and coding using each prop in the prop sequence. Specifically, a target prop code is input into a first full-connection network, and first intermediate result information is output through calculation of each parameter, wherein the first intermediate result information comprises an intermediate vector of the target prop; and inputting the comparison prop code into a second full-connection network, and outputting second intermediate result information through calculation of each parameter, wherein the second intermediate result information comprises an intermediate vector of the comparison prop.

Step B2, inputting the initial property information of the property using sequence into the first sequence network, and outputting third intermediate information;

the implementation method of the first sequence Network may be various, for example, models such as 1D CNN (One-dimensional convolutional Neural Networks), RNN (Recurrent Neural Networks), and Transformer (open source deep learning model of google), where the Transformer is mainly applied in the field of NLP (Natural Language Processing) and excels in Processing data of sequence type; specifically, the used prop codes included in the used prop sequence are input into the first sequence network, and third intermediate result information is output through calculation of each parameter, wherein the third intermediate result information includes intermediate vectors of each used prop.

Step B3, inputting the first intermediate information, the second intermediate information and the third intermediate information into a third fully-connected network to obtain a first output result, and determining a target prop and second sequencing information of the comparison prop based on the first output result; the second ranking information includes: the probability that the target prop is arranged in front of the comparison prop.

Specifically, the first intermediate information, the second intermediate information and the third intermediate information may be spliced, the splicing result is input to a third fully-connected network, and a first output result is obtained through calculation of each parameter in the third fully-connected network, where the first output result includes probability information that the target prop is arranged in front of the comparison prop, the probability information is input to a Sigmoid network, and the probability information may be mapped to an interval of (0, 1) to represent a probability that a target condition is satisfied, that is, a probability that the target prop is arranged in front of the comparison prop may be obtained.

It should be noted that, by training the first auxiliary model, the first auxiliary model can rank the precedence relationship of the props in the backpack. However, the first auxiliary model uses future data, which means that a training sample is generated for each movement (exchange) of the prop, and the future data is a use prop sequence corresponding to the movement (exchange) time, that is, a use prop sequence of a user after the time; therefore, the future data cannot be acquired when the online service is performed, so that the first auxiliary model cannot be directly applied to the online service.

Step S604, inputting each prop in the training sample into a middle training result, and outputting middle prop information of each prop;

the intermediate training result is an information coding model after the property sequential training is completed, in order to continue training the coding mode of the information coding model, so that the information coding model after the training can distinguish different positions in a backpack and measure the adjacent relation of properties, each property in a training sample can be input into the intermediate training result, and the intermediate property information of each property is output, wherein the intermediate property information comprises target property coding, comparison property coding and property sequence coding.

Step S606, inputting the middle prop information of each prop into a second auxiliary model, and outputting position information indicating whether the target prop and the comparison prop are adjacent; and training the information encoder and the second auxiliary model based on the first label information and the position information until the second auxiliary model converges to obtain the trained information encoder.

The second auxiliary model may be various Network models, such as CNN (Convolutional Neural networks), RNN (Recurrent Neural networks), DNN (deep Neural networks), and the like; the second auxiliary model is typically a model that has not been trained; in addition, when the order of the props at a certain moment is predicted, the data of the props used after the moment are utilized, and the second auxiliary model can also be called a posterior prop adjacent model; therefore, the first tag information in the intermediate item information of each item input to the second auxiliary model is the adjacent relation between the target item and the comparison item, and can be represented by "2" and "3". The upper position information indicating whether the target prop and the comparison prop are adjacent to each other may be adjacent probability of the target prop and the comparison prop, or specific position information of the target prop and the comparison track, for example, the target prop is adjacent to the comparison prop, or the target prop is not adjacent to the comparison prop.

Specifically, the first label information is compared with the position information output by the second auxiliary model, and learnable parameters in the second auxiliary model and the information coding model are trained by using a back propagation technique and the like, wherein the learnable parameters include: scientific system parameters in the information coding model and the second auxiliary model are determined until the second auxiliary model converges, so that the second auxiliary model can judge the adjacent relation of the props in the backpack; after the second auxiliary model is converged, the trained information encoder can be obtained, and when the second auxiliary model is trained, the information encoding model can learn an optimal encoding mode from data of adjacent relations of the props; by the coding mode, the adjacent relation of the props can be measured. The more similar the prop is encoded, the greater the likelihood of its proximity and the greater the likelihood of the user using them simultaneously. For example, for user a, under game scenario B, props that approximate the encoding of user a and scenario B tend to be used.

In addition, the second auxiliary model convergence may be that the performance of the second auxiliary model is not significantly different before and after training, i.e. the second auxiliary model is considered to have converged; the second auxiliary model can also be set with a loss value, and when the loss value output for each input data is fixed or very close to a certain fixed value, the model is considered to be converged; the condition that the loss value satisfies a preset loss threshold may be further used as a condition for convergence of the second auxiliary model.

It should be noted that, in the actual execution process of the steps S602 and S606, the order is not limited, and the process of inputting the initial item information of each item into the second auxiliary model in the step S606 may be executed first.

In a preferred embodiment, the second auxiliary model includes: a fourth fully connected network, a fifth fully connected network, a second sequence network, and a sixth fully connected network; the step of inputting the intermediate prop information of each prop into the second auxiliary model and outputting the position information of whether the target prop and the comparison prop are adjacent comprises the following steps:

step C1, inputting the intermediate prop information of the target prop into a fourth fully-connected network, and outputting fourth intermediate information; inputting the intermediate prop information of the comparison props into a fifth fully-connected network, and outputting fifth intermediate information;

referring to fig. 8, a flow chart of a training method of a second auxiliary model is shown, in which parameters in an encoder of intermediate training results of a fourth fully-connected network, a fifth fully-connected network and an information coding model need to be learned. In actual implementation, the intermediate training result and the second auxiliary model may be spliced, all the prop quadruples in the input data are encoded by using the information encoding model that has completed the prop sequential training, that is, the intermediate training result, and then the output result of the intermediate training result, that is, the intermediate prop information, is used as the input of the second auxiliary model, which includes target prop encoding, comparison prop encoding, and encoding using each prop in the prop sequence. Specifically, the target prop code is input into a fourth fully-connected network, and fourth intermediate result information is output through calculation of each parameter, wherein the fourth intermediate result information comprises an intermediate vector of the target prop; and inputting the comparison prop code into a fifth full-connection network, and outputting fifth intermediate result information through calculation of each parameter, wherein the fifth intermediate result information comprises an intermediate vector of the comparison prop.

Step C2, inputting the intermediate item information using the item sequence into the second sequence network, and outputting sixth intermediate information;

the implementation method of the second sequence Network may be various, for example, models such as 1D CNN (One-dimensional convolutional Neural Networks), RNN (Recurrent Neural Networks), and Transformer (open source deep learning model of google), where the Transformer is mainly applied in the field of NLP (Natural Language Processing) and excels in Processing data of sequence type; specifically, the used item codes included in the used item information of the item sequence are input into the second sequence network, and sixth intermediate result information is output through calculation of each parameter, wherein the sixth intermediate result information includes intermediate vectors of each used item.

Step C3, inputting the fourth intermediate information, the fifth intermediate information and the sixth intermediate information into a sixth fully-connected network to obtain a second output result, and determining whether the position information indicating the target prop and the comparison prop are adjacent or not based on the second output result; the location information includes: and the probability that the target prop is adjacent to the comparison prop.

Specifically, the fourth intermediate information, the fifth intermediate information and the sixth intermediate information may be spliced, the splicing result is input to a sixth fully-connected network, a second output result is obtained through calculation of each parameter in the sixth fully-connected network, the second output result includes probability information that the target prop is adjacent to the comparison prop, the probability information is input to a Sigmoid network, the probability information may be mapped to an interval of (0, 1), a probability that the target condition is met is represented, and then the probability that the target prop is adjacent to the comparison prop can be obtained.

It should be noted that, by training the second auxiliary model, the second auxiliary model can learn the adjacent relationship of the props in the backpack. However, the second auxiliary model uses future data, where the future data is a training sample generated for each movement (exchange) of the prop, and the future data is a prop-using sequence corresponding to the movement (exchange) time, that is, a prop-using sequence of the user after the time; therefore, the future data cannot be acquired when the online service is performed, so that the second auxiliary model cannot be directly applied to the online service.

In the method, a first auxiliary model and an information coding model are trained through a training sample to obtain an intermediate training result of the information coding model, then the training sample is input into the intermediate training result, a second auxiliary model and the information coding model are trained by utilizing output intermediate prop information, and the purpose is to enable the information coding model to obtain an optimal coding mode, wherein the first auxiliary model can enable the information coding model to distinguish the sequence of props; the second auxiliary model can enable the information coding model to have the adjacent relation for measuring props, and the first auxiliary model and the second auxiliary model are trained, so that the information coding model can have the two functions at the same time; the method does not depend on game plans to manually configure the property codes, but automatically learns the position relation of the properties through a machine learning model, thereby reducing the pressure of the game plans and simultaneously avoiding adverse consequences caused by wrong configuration of the property codes; in the mode, the use stage of the information coding model can also aim at the continuous storage habits of the users on different props for the users, and aim at different users to carry out personalized arrangement on the props, so that the personalized requirements of different players are met.

The embodiment also provides another method for arranging game props, which describes a specific implementation manner of the steps of inputting prop information of a first prop and prop information of a second prop in a prop pair into a pre-trained prop sequence model and outputting first sequence information of the first prop and the second prop, wherein the prop sequence model comprises a seventh fully-connected network, an eighth fully-connected network and a ninth fully-connected network; the method specifically comprises the following steps:

step D1, inputting the prop information of the first prop into a seventh fully-connected network, and outputting seventh intermediate information; inputting prop information of the second prop to an eighth fully-connected network, and outputting eighth intermediate information;

the parameters included in the seventh fully connected network, the eighth fully connected network and the ninth fully connected network are networks trained in advance; the property information of the first property comprises a code corresponding to a user in a scene, the code is input into a seventh fully-connected network, and seventh intermediate information is output through calculation of each parameter in the seventh fully-connected network, wherein the seventh intermediate information comprises an intermediate vector of the first property; and the prop information of the second prop comprises a code of the second prop relative to the user in the scene, the code is input into an eighth fully-connected network, and eighth intermediate information is output through calculation of various parameters in the eighth fully-connected network, wherein the eighth intermediate information comprises an intermediate vector of the second prop.

Step D2, inputting the seventh intermediate information and the eighth intermediate information into the ninth fully-connected network to obtain a third output result; determining first sequencing information of the first prop and the second prop based on the third output result; wherein the first ordering information includes: the first prop has a probability of being arranged in front of the second prop, or the result of the ordering of the first prop and the second prop.

Specifically, the seventh intermediate information and the eighth intermediate information are spliced and input to a ninth fully-connected network, and a third output result is obtained through calculation of each parameter in the ninth fully-connected network, where the output result may include probability information that the first item is arranged in front of the second item, the probability information is input to a Sigmoid network, and the probability information may be mapped to a (0, 1) interval to represent a probability that meets a target condition, that is, a probability that the first item is arranged in front of the second item is obtained; or the output result can also comprise a sequencing result of the first prop and the second prop, and the sequencing order of the first prop and the second prop can be determined directly according to the sequencing result.

Referring to a flow diagram of a training method of a property sequence model shown in fig. 9, the property sequence model is obtained by training through the following steps:

step E1, obtaining sample prop pairs and second label information; the sample prop pair comprises a target prop and a comparison prop in the sample prop set; wherein, the target prop is a prop which has changed position; the second tag information includes: the front-back position relation between the target prop and the comparison prop;

the second tag information may be represented by a numerical value, for example, the tag "1" is used to represent the target prop in the front position of the comparison prop; and the target prop is positioned behind the comparison prop and is represented by a label '0'. In actual implementation, the target prop and the comparison prop are coded through the trained information coding model, so that the code of the target prop relative to the user in the scene is obtained, and the code of the comparison prop relative to the user in the scene is obtained.

Step E2, inputting the prop information of the target prop into a seventh fully-connected network, and outputting seventh intermediate information; inputting the prop information of the comparison props into an eighth fully-connected network, and outputting eighth intermediate information;

inputting the prop information of the target prop, namely the code of the target prop relative to the user in the scene, into a seventh fully-connected network, and outputting seventh intermediate result information through calculation of each parameter, wherein the seventh intermediate result information comprises an intermediate vector of the target prop; inputting the prop information of the comparison props into an eighth fully-connected network, and outputting eighth intermediate result information through calculation of each parameter, wherein the eighth intermediate result information comprises intermediate vectors of the comparison props.

Step E3, inputting the seventh intermediate information and the eighth intermediate information into a ninth fully-connected network to obtain a third output result, and obtaining a target prop and third sequencing information in front of the comparison prop based on the third output result;

the third sorting information may be probability information that the target prop is arranged in front of the comparison prop, or may be a sorting result of the target prop and the comparison prop. Specifically, the seventh intermediate information and the eighth intermediate information are spliced, the splicing result is input to a ninth fully-connected network, a third output result is obtained through calculation of each parameter in the ninth fully-connected network, the third output result comprises a target prop and third sequencing information in front of a comparison prop, namely probability information that the target prop is arranged in front of the comparison prop, or the probability information is input to a Sigmoid network according to the sequencing result of the target prop and the comparison prop, the probability information can be mapped to an interval of (0, 1), the probability meeting the target condition is represented, and the probability that the target prop is arranged in front of the comparison prop can be obtained.

And E4, training the prop sequence model based on the second label information and the third sequencing information until the prop sequence model converges, and obtaining a trained prop training model.

Comparing the second label information with the third sequencing information, and training learnable parameters in the prop sequence model by using a back propagation technology, wherein the learnable parameters comprise all parameters in a seventh fully-connected network, an eighth fully-connected network and a ninth fully-connected network; and obtaining the trained property training model until the property sequence model converges. The trained prop sequence model can be used in an online service module.

In addition, the convergence of the prop sequence model can be that the performance of the prop sequence model is not obviously different before and after training, namely the prop sequence model is considered to be converged; the property sequence model can also be set with a loss value, and when the output loss value is fixed or very close to a certain fixed value for data input each time, the model is considered to be converged; and the loss value can also meet a preset loss threshold value as a convergence condition of the prop sequence model.

In the method, a prop sequence model is trained through a trained information coding model, so that the prop sequence model can identify the position relation of props; the method can automatically learn the precedence relationship of the props through a machine learning model without depending on game plans to manually configure the prop codes. The pressure of game planning is reduced, and adverse effects caused by wrong configuration of the property codes are avoided; can learn the position relation between the stage property through stage property order model, according to the user to the storage custom of different stage properties, carry out individualized arrangement to user's stage property, need not to sort in order according to the serial number of stage property, improved the flexibility of the arrangement order of stage property in the knapsack, make the arrangement order of stage property and player's individual custom and the actual demand phase-match of recreation scene simultaneously, improved player's recreation experience.

The embodiment also provides another method for arranging game props, and the method describes a specific implementation manner of a step of acquiring at least one pair of prop pairs from a prop set and a specific implementation manner of a step of ordering the props in the prop set according to first ordering information corresponding to each pair of prop pairs;

first, a specific implementation manner of the step of obtaining at least one pair of prop pairs from the prop set is described, which includes: clustering the props in the prop set to obtain at least one prop cluster; each prop type cluster comprises at least one prop in a prop set; aiming at each prop type cluster, if the current prop type cluster comprises a plurality of props, generating at least one pair of prop pairs based on the plurality of props;

the clustering processing can be used for clustering the props in the prop set through a clustering model; the Clustering model may be a Clustering algorithm (Clustering by fast search and find of diversity peaks) based on density peaks, which is a mature Clustering algorithm. Therefore, the clustering model does not need to be subjected to parameter training and can be directly classified in the online service module according to the code of the prop. The props included in each prop cluster may be coded similar props, uncommon props, props with similar functions or used by players frequently at the same time, and the like. Specifically, the props in the prop set can be encoded by an information encoder, then clustering is carried out by a density peak value clustering algorithm in a clustering model, and props with similar codes can be clustered into a cluster; for example, as a result of the clustering process shown in fig. 10, the props are divided into 4 clusters. The number of the props in each cluster can be different, wherein the prop a, the prop f, the prop h and the prop i are respectively the center of each cluster.

Secondly, describing the step of sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs, including: counting the times of each prop in the current prop cluster being arranged in front of each prop pair according to the first sequencing information corresponding to each pair of prop in the current prop cluster; and sequencing the props in the current prop type cluster based on the number of times that each prop is arranged in front of each prop pair.

Specifically, all the props in each prop cluster are paired pairwise, and prop information is input into the prop sequence model, so that the sequential relation of any two props in each prop cluster can be obtained. Counting the times of each prop in the current prop cluster in front of other props, sequencing all props in the current prop cluster according to the times, and obtaining the ranking of each prop in the cluster, wherein the prop with the first ranking is used as a representative in the cluster.

In addition, similar or frequently used props can be put together by means of clustering.

In the method, props with similar functions or frequently used by players at the same time can be put together in a clustering way; the game experience of the player when the stage property is continuously used is improved; meanwhile, the properties which are not frequently used can be automatically identified, and the properties are put together and placed at the position close to the back of the backpack, so that the effective backpack space can be more effectively utilized; the properties are not required to be sequenced according to the numbers of the properties, so that the flexibility of the sequence of the properties in the backpack is improved, the sequence of the properties is matched with the personal habits of the players and the actual requirements of the game scenes, and the game experience of the players is improved.

This implementation also provides that after the step of ranking the props in the current prop category cluster based on the number of times each prop is ranked ahead in each prop pair, the method further includes the steps of:

step F1, setting the current scene information in the prop information of each prop in the current prop cluster to zero to obtain the temporary prop information of each prop;

in the actual game process, the game is in different scenes, and players have different requirements on property arrangement; for example, when a monster is played in the copy, the player has a relatively large demand for basic drugs; when the copy is played with BOSS, the player has great demands for advanced drugs and gain drugs; when the player forges the equipment, the requirements on minerals, design drawings and the like are high; therefore, the props can be highlighted according to the characteristics of the current scene in the process of arranging the props by one key; so that the player can quickly obtain the props required by the current scene.

In actual implementation, in order to obtain the props related to the current scene, current scene information in the prop information of each prop in the current prop cluster may be set to zero, so as to obtain temporary prop information of each prop.

Step F2, inputting the temporary prop information of the first prop and the temporary prop information of the second prop in the prop pair to a prop sequence model and outputting the temporary arrangement information of the first prop and the second prop aiming at each pair of prop pairs in the current prop cluster;

step F3, sorting the props in the prop set according to the temporary sorting information corresponding to each pair of props to obtain a temporary sorting result;

and step F4, determining the props related to the current scene in the current props cluster based on the temporary sorting result.

Specifically, the process is similar to the process of online arrangement of props, and is not repeated again, after the final temporary sequencing result is obtained, the temporary sequencing result can be compared with the sequencing result of the original prop type cluster, and props with the number difference between the ranking digit number of the temporary sequencing result in the cluster and the ranking digit number of the sequencing result of the original prop type cluster exceeding 3 are considered as props related to the scene.

In the method, the game scene of the current user is identified in a mode that the current scene information in the prop information of each prop is set to be zero, the props relevant to the current scene are determined by the online service module and are highlighted, the time required for the user to search the props relevant to the scene is reduced, and the game experience of the user is improved.

Further, after finishing ranking of the props in the prop cluster, the method further comprises:

determining the props which are arranged in front of all prop pairs in the current prop cluster most frequently as representative props in the current prop cluster aiming at each prop cluster; generating at least one pair of representative prop pairs based on the representative props of each prop type cluster; counting the times of each representative prop being arranged in front of each representative prop pair according to the first sequencing information corresponding to each representative prop pair; each prop category cluster is ranked based on the number of times each representative prop is ranked first in each representative prop pair.

For example, referring to the clustering result shown in fig. 10, the props with the most times arranged in front of each prop pair in the front prop cluster, that is, the representative props in the four prop clusters in the figure are prop a, prop f, prop h, and prop i, respectively; based on the four representative props, generating one pair or a plurality of pairs of representative prop pairs, which can be prop a and prop f, prop a and prop i, prop a and prop h, prop f and prop h, and the like, inputting the prop information of each prop pair into a prop sequence model, determining the times of the prop a in front of other props, the times of the prop f in front of other props, the times of the prop h in front of other props, and the times of the prop i in front of other props according to the output probability, and sequencing the prop cluster where the prop a is located, the prop cluster where the prop f is located, the prop cluster where the prop h is located, and the prop cluster where the prop i is located according to the times in the sequence from large to small.

Further, after the step of sorting the props in the prop set according to the first sorting information corresponding to each pair of prop pairs, the method further includes: responding to the prop adjusting instruction, and acquiring a prop to be adjusted and a target position corresponding to the prop adjusting instruction; and adjusting the prop to be adjusted in the prop set to a target position.

The prop to be adjusted can be any prop in a backpack; in actual implementation, the props after arrangement are displayed in response to the prop arrangement instruction, a fine adjustment interface is provided, the props to be adjusted in the prop set can be adjusted according to the own will of the player according to the props to be adjusted and the target position of the props to be adjusted in the interface. The specific interaction process comprises the steps that the client receives a prop sequence returned by the server, the prop sequence is displayed on a game interface, a fine-tuning interface is provided, and a player can adjust the placement of props in the fine-tuning interface; the client receives the 'adjustment' request, the client sends a request for recording 'adjustment', and the server records the player adjustment record and returns a response; and the client receives the response of the server and displays the adjusted prop.

In the method, the prop sequence model is used for sequencing the prop cluster pieces by using the representative props of the prop cluster, the props in the backpack can be sequenced according to the containing habits of users, the sequenced props are displayed, and a game interface convenient for the users to adjust is provided, so that the users can adjust the one-key arrangement result; the properties are not required to be sequenced according to the numbers of the properties, so that the flexibility of the sequence of the properties in the backpack is improved, the sequence of the properties is matched with the personal habits of the players and the actual requirements of the game scenes, and the game experience of the players is improved.

The embodiment provides a specific timing chart of a method for arranging game props, and as shown in fig. 11, the scheme for arranging the props mainly comprises three parts: the system comprises a data module, a model training module and an online service module. Wherein, the M0 model corresponds to the information coding model; the M1 model corresponds to the first auxiliary model; the M2 model corresponds to the aforementioned second auxiliary model; the M3 model corresponds to the clustering model; the M4 model corresponds to the aforementioned prop sequence model.

The data module mainly comprises three steps, namely first off-line data collection, wherein a property using sequence of a player, user portrait information and property moving data of the player are obtained from a log database, wherein the property using sequence comprises a moved property, a position before moving, a position after moving, a scene during moving and backpack data of the player, and the position of each property in a backpack is included. Second data preprocessing, wherein the process processes the data into data which can be understood by a network model corresponding to the prop information generation mode; the third constructed training data set corresponds to the aforementioned training sample, wherein the training data of the M1, M2 and M4 models are slightly different, and the labels of the data are different. Training data sets used by the M1 and M4 models are prop sequence data sets; the training data set used by the M2 model is the prop neighbor data set.

The model training module is mainly divided into grid steps, wherein in the first step, M0: encoder model and M1: splicing the posterior prop sequence models, and training the two models by using a prop sequence data set; second, M0: encoder model and M2: splicing the adjacent models of the posterior properties, and training the two models by using the adjacent data sets of the properties; thirdly, judging whether the M1 model and the M2 model have converged, if so, entering the next step, otherwise, returning to the first step; the fourth step, taking out the M1 model and the M0 model in the M2 model, and the fifth step, taking out the M0: encoder model and M4: and splicing the prop sequence models, training the M4 model by using the prop sequence data set, and not training the M0 model at the moment.

The online service module 1, a player clicks a 'one-key sorting' button; 2. the client receives a request of clicking 'one-key sorting' by a player; 3. a client sends a 'one-key arrangement' request to a server, and parameters are a backpack prop sequence, a current scene and a user image; 4. the server acquires parameters, processes all props in the backpack prop sequence into a quadruple, and codes the quadruple by using an M0 model to obtain a prop code; 5. calling an M3 model, and clustering the props in the backpack according to the prop codes to obtain k clusters; 6. calling an M4 model, and obtaining the sequencing of props in each cluster, the sequencing among clusters and the highlighted props; the highlighted prop identifies a prop associated with the current scene. Returning the finished prop sequence; 7. the client receives the prop sequence returned by the server, displays the prop sequence on a game interface and provides a fine-tuning interface; 8. the player adjusts the placement of the props in the fine adjustment interface; 9. the client receives a request for 'adjustment'; 10. the client sends a request for recording 'adjustment'; 11. the server records the player adjustment record and returns a response; 12. and the client receives the response of the server and displays the adjusted prop.

The specific method for arranging the game props provided by the embodiment has the same technical characteristics as the method for arranging the game props provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.

Corresponding to the above method embodiment, referring to fig. 12, a schematic structural diagram of an organizing device of a game item is shown, the device includes:

the property collection obtaining module 121 is configured to, in response to a property arrangement instruction, obtain a property collection of a player corresponding to the property arrangement instruction;

a prop pair obtaining module 122, configured to obtain at least one pair of prop pairs from the prop set;

the sequencing information output model 123 is configured to, for each pair of prop pairs, input prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and output first sequencing information of the first prop and the second prop; wherein, the prop information is determined according to the current game scene and/or the current player;

and the prop sequencing module 124 is configured to sequence the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs.

The embodiment of the invention provides a game prop arrangement device, which responds to a prop arrangement instruction and acquires a prop set of a player corresponding to the prop arrangement instruction; then at least one pair of prop pairs is obtained from the prop set; for each pair of prop pairs, inputting prop information of a first prop and prop information of a second prop in the prop pairs into a pre-trained prop sequence model, and outputting first sequence information of the first prop and the second prop; and sequencing the props in the prop set according to the first sequencing information corresponding to each pair of prop pairs. In the mode, the order relation of each prop in the backpack is automatically learned through the prop order model, the props in the player backpack are sequenced based on the order relation, sequencing is not needed according to the serial number of the props, the flexibility of the sequence of the props in the backpack is improved, meanwhile, the sequence of the props is matched with the personal habits of the players and the actual requirements of game scenes, and the game experience degree of the players is improved.

Further, the device also comprises a prop information obtaining module, which is used for obtaining the original information of the prop to be processed; the original information comprises a prop identification, a prop type, current game scene information and current player information of the prop to be processed; and inputting the original information into an information encoder which is trained in advance to obtain the prop information of the prop to be processed.

Further, the device also comprises an information encoder training module, which is used for obtaining a training sample containing a plurality of props; inputting each prop in the training sample into an information encoder, and outputting initial prop information of each prop; and inputting the initial prop information of each prop into an auxiliary training model so as to train the information encoder based on the auxiliary training model to obtain the trained information encoder.

Further, the training samples include: the method comprises the steps of obtaining a target prop, a comparison prop, first tag information and a use prop sequence in a sample prop set; wherein, the target prop is a prop which has changed position; the first tag information includes: the front-back position relation between the target prop and the comparison prop and whether the target prop and the comparison prop have an adjacent relation; the using prop sequence is as follows: the method comprises the steps that after a position change occurs to a target prop, a prop sequence used by a player in a sample prop set is obtained; the props in the prop use sequence are arranged according to the used time sequence.

Further, the auxiliary training model comprises a first auxiliary model and a second auxiliary model; the information encoder training module is further configured to: inputting the initial prop information of each prop into a first auxiliary model, and outputting a target prop and second arrangement information of the comparison props; training the information encoder and the first auxiliary model based on the first label information and the second arrangement information until the first auxiliary model converges to obtain an intermediate training result of the information encoder; inputting each prop in the training sample into an intermediate training result, and outputting intermediate prop information of each prop; inputting the middle prop information of each prop into a second auxiliary model, and outputting position information indicating whether the target prop and the comparison prop are adjacent or not; and training the information encoder and the second auxiliary model based on the first label information and the position information until the second auxiliary model converges to obtain the trained information encoder.

Further, the first auxiliary model includes: a first fully connected network, a second fully connected network, a first sequence network, and a third fully connected network; the information encoder training module is further configured to: inputting initial prop information of a target prop into a first fully-connected network, and outputting first intermediate information; inputting the initial prop information of the comparison props into a second fully-connected network, and outputting second intermediate information; inputting initial prop information using a prop sequence into the first sequence network, and outputting third intermediate information; inputting the first intermediate information, the second intermediate information and the third intermediate information into a third fully-connected network to obtain a first output result, and determining a target prop and comparing second sequencing information of the props based on the first output result; the second ranking information includes: the probability that the target prop is arranged in front of the comparison prop.

Further, the second auxiliary model includes: a fourth fully connected network, a fifth fully connected network, a second sequence network, and a sixth fully connected network; the information encoder training module is further configured to: inputting the intermediate prop information of the target prop into a fourth fully-connected network, and outputting fourth intermediate information; inputting the intermediate prop information of the comparison props into a fifth fully-connected network, and outputting fifth intermediate information; inputting the intermediate prop information using the prop sequence into a second sequence network, and outputting sixth intermediate information; inputting the fourth intermediate information, the fifth intermediate information and the sixth intermediate information into a sixth fully-connected network to obtain a second output result, and determining position information indicating whether the target prop and the comparison prop are adjacent or not based on the second output result; the location information includes: and the probability that the target prop is adjacent to the comparison prop.

Further, the prop sequence model comprises a seventh fully connected network, an eighth fully connected network and a ninth fully connected network; the ranking information output model is further configured to: inputting the prop information of the first tool to a seventh fully-connected network, and outputting seventh intermediate information; inputting prop information of the second prop to an eighth fully-connected network, and outputting eighth intermediate information; inputting the seventh intermediate information and the eighth intermediate information to a ninth fully-connected network to obtain a third output result; determining first sequencing information of the first prop and the second prop based on the third output result; wherein the first ordering information includes: the first prop has a probability of being arranged in front of the second prop, or the result of the ordering of the first prop and the second prop.

Further, the device also comprises a prop sequence model training module, which is used for acquiring the sample prop pair and the second label information; the sample prop pair comprises a target prop and a comparison prop in the sample prop set; wherein, the target prop is a prop which has changed position; the second tag information includes: the front-back position relation between the target prop and the comparison prop; inputting prop information of the target prop into a seventh fully-connected network, and outputting seventh intermediate information; inputting the prop information of the comparison props into an eighth fully-connected network, and outputting eighth intermediate information; inputting the seventh intermediate information and the eighth intermediate information into a ninth fully-connected network to obtain a third output result, and obtaining a target prop and third sequencing information adjacent to the comparison prop based on the third output result; and training the prop sequence model based on the second label information and the third sequencing information until the prop sequence model converges to obtain a trained prop training model.

Further, the prop pair obtaining module is further configured to: clustering the props in the prop set to obtain at least one prop cluster; each prop type cluster comprises at least one prop in a prop set; aiming at each prop type cluster, if the current prop type cluster comprises a plurality of props, generating at least one pair of prop pairs based on the plurality of props; the prop sequencing module is further configured to: counting the times of each prop in the current prop cluster being arranged in front of each prop pair according to the first sequencing information corresponding to each pair of prop in the current prop cluster; and sequencing the props in the current prop type cluster based on the number of times that each prop is arranged in front of each prop pair.

Further, the above apparatus is further configured to: setting current scene information in the prop information of each prop in the current prop cluster to be zero, and obtaining temporary prop information of each prop; aiming at each pair of prop pairs in the current prop cluster, inputting the temporary prop information of a first prop and the temporary prop information of a second prop in the prop pairs into a prop sequence model, and outputting the temporary sequencing information of the first prop and the second prop; sequencing the props in the prop set according to the temporary sequencing information corresponding to each pair of props to obtain a temporary sequencing result; and determining the props related to the current scene in the current props cluster based on the temporary sequencing result.

Further, the above apparatus is further configured to: determining the props which are arranged in front of all prop pairs in the current prop cluster most frequently as representative props in the current prop cluster aiming at each prop cluster; generating at least one pair of representative prop pairs based on the representative props of each prop type cluster; counting the times of each representative prop being arranged in front of each representative prop pair according to the first sequencing information corresponding to each representative prop pair; each prop category cluster is ranked based on the number of times each representative prop is ranked first in each representative prop pair.

Further, the above apparatus is further configured to: responding to the prop adjusting instruction, and acquiring a prop to be adjusted and a target position corresponding to the prop adjusting instruction; and adjusting the prop to be adjusted in the prop set to a target position.

The game prop arrangement device provided by the embodiment of the invention has the same technical characteristics as the game prop arrangement method provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved.

The embodiment also provides electronic equipment, which comprises a processor and a memory, wherein the memory stores machine executable instructions capable of being executed by the processor, and the processor executes the machine executable instructions to realize the method for arranging the game props.

Referring to fig. 13, the electronic device includes a processor 100 and a memory 101, where the memory 101 stores machine executable instructions capable of being executed by the processor 100, and the processor 100 executes the machine executable instructions to implement the method for organizing the game items.

Further, the electronic device shown in fig. 13 further includes a bus 102 and a communication interface 103, and the processor 100, the communication interface 103, and the memory 101 are connected by the bus 102.

The Memory 101 may include a high-speed Random Access Memory (RAM) and may also include a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. The communication connection between the network element of the system and at least one other network element is realized through at least one communication interface 103 (which may be wired or wireless), and the internet, a wide area network, a local network, a metropolitan area network, and the like can be used. The bus 102 may be an ISA bus, PCI bus, EISA bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one double-headed arrow is shown in FIG. 13, but that does not indicate only one bus or one type of bus.

Processor 100 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware or instructions in the form of software in the processor 100. The Processor 100 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the device can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component. The various methods, steps and logic blocks disclosed in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present invention may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in the memory 101, and the processor 100 reads the information in the memory 101 and completes the steps of the method of the foregoing embodiment in combination with the hardware thereof.

The present embodiments also provide a machine-readable storage medium having stored thereon machine-executable instructions that, when invoked and executed by a processor, cause the processor to implement the above-described method of organizing play objects.

The method and the device for organizing game items and the computer program product of the electronic device provided by the embodiment of the invention comprise a computer readable storage medium storing program codes, wherein instructions included in the program codes can be used for executing the method described in the previous method embodiment, and specific implementation can refer to the method embodiment, and is not described herein again.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the system and the apparatus described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In addition, in the description of the embodiments of the present invention, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the present invention can be understood in specific cases for those skilled in the art.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that the following embodiments are merely illustrative of the present invention, and not restrictive, and the scope of the present invention is not limited thereto: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.