阅读说明：本技术 一种基于目标节目推荐模型的节目推荐方法与装置 (Program recommendation method and device based on target program recommendation model ) 是由 陈飞鸽 廖佳秋 于 2021-09-29 设计创作，主要内容包括：本发明提供了一种基于目标节目推荐模型的节目推荐方法,通过提取第一节目向量,计算所述第一节目向量与各个节目推荐模型预设的推荐向量之间的相似度,并进行相似度比较,根据比较结果选取目标节目推荐模型,并将第一节目向量输入至目标节目推荐模型中,得到多个第二节目,将各所述第二节目推送给所述指定设备。从而实现了无需获取到用户的用户信息,直接根据用户当前播放的第一节目,就能完成节目的推荐,并且,还通过获取相关的推荐模型,提高了节目推荐的精确度,解决了在没有确定数字电视面前的人员时,很难为该人员推荐相关的电视节目的问题。(The invention provides a program recommendation method based on a target program recommendation model, which comprises the steps of extracting a first program vector, calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model, comparing the similarity, selecting the target program recommendation model according to the comparison result, inputting the first program vector into the target program recommendation model to obtain a plurality of second programs, and pushing each second program to designated equipment. Therefore, the program recommendation can be completed directly according to the first program currently played by the user without acquiring the user information of the user, the program recommendation accuracy is improved by acquiring the related recommendation model, and the problem that related television programs are difficult to recommend to people before the digital television is not determined is solved.)

1. A program recommendation method based on a target program recommendation model is characterized by comprising the following steps:

extracting a first program type and first program information of a first program currently played by a designated device;

vectorizing the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

summing the first type vector and the first information vector according to a preset weighted summing mode to obtain a first program vector;

acquiring second program types and second program information of a plurality of programs played by a plurality of devices respectively;

vectorizing the second program type and the second program information to obtain a corresponding second type vector and a corresponding second information vector;

summing the second type vector and the second information vector corresponding to each device according to a preset weighted summation mode to obtain a second program vector;

clustering each second program vector according to preset dimensions to obtain a training set corresponding to each preset dimension;

combining every two second program vectors in the training set in pairs and sequentially inputting the combined second program vectors into an input layer of an initial model corresponding to a preset dimension by adopting a formula v_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model;

according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein Rkpq is the p second program vector and the q second program vector in the k training setSimilarity values corresponding to the program vectors;

in the discrimination network of the initial model, according to a formula

Calculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

adjusting parameters of the initial model based on the loss values corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimension after adjustment;

calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model, and comparing the similarity;

selecting a target program recommendation model with the highest similarity according to the comparison result, and inputting the first program vector into the target program recommendation model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program;

and pushing each second program to the designated equipment.

2. The method of claim 1, wherein the step of extracting the first program information from the first program information and the first program type of the first program currently played by the specific device comprises:

acquiring a search word of the designated device based on the first program, and determining a related word set of the search word, wherein the related word set comprises one or more related words;

determining one or more target related words from the related word set according to the first part of speech of the search word and the second part of speech of each related word;

determining a plurality of reference devices according to the search term and the one or more target related terms; wherein the reference device is a device that searches the first program using the search term and at least one of the one or more target related terms;

and determining common characteristics of the reference devices, and recording the common characteristics as the first program information.

3. The method of claim 1, wherein before the step of pushing each of the second programs to the specific device, the method further comprises:

acquiring target program vectors corresponding to a plurality of target programs of which the playing time of the appointed equipment is greater than a set value; wherein the target program vector comprises a program type of the target program and the program information;

extracting the dimension value of each dimension in each target program vector from preset dimensions;

extracting dimension values of all dimensions corresponding to all target program vectors according to all the preset dimensions respectively to obtain a dimension value set corresponding to all the preset dimensions respectively;

extracting the maximum value and the minimum value in each dimension set;

according to the formulaCalculating a standard value of a dimension value corresponding to each dimension set, wherein x is_ijRepresents the ith dimension value, min (x) in the jth dimension value data set_ij) Represents the minimum value, max (x), of the element in the jth of the dimension numerical data set_ij) Maximum value, Y, of an element in jth of the dimension numerical data set_ijRepresenting the standard value corresponding to the ith dimension value in the dimension value data set;

according to the formulaCalculating information entropy values for each of the sets of dimensions, whereinWherein E is_jSaid information entropy value representing the jth of said dimension set, when P_ijWhen =0, defineRepresenting a probability value corresponding to the ith dimension value of the jth dimension set, wherein n represents the number of the dimension sets;

comparing the information entropy value of each dimensionality set with preset information entropy values corresponding to each preset dimensionality;

screening out the dimension set smaller than the corresponding preset information entropy value according to a comparison result, and recording as a target dimension set;

recording the minimum value in each target dimension set as a dimension requirement value corresponding to a preset dimension;

obtaining a dimension value of each second program corresponding to the dimension of the target dimension set, and judging whether the dimension value is greater than the corresponding dimension requirement value;

and if the dimension requirement value is larger than the corresponding dimension requirement value, judging that the corresponding second program meets the requirement of pushing the second program to the specified equipment.

4. The method of claim 1, wherein after the step of pushing each of the second programs to the designated device, the method further comprises:

acquiring a second program selected by a user on the designated equipment, and recording the second program as a second target program;

extracting program information and playing integrity of each second target program;

converting the integrity degree into a corresponding recommended score according to a preset score conversion method;

and using the program information as the input of the target recommendation model, using the recommendation score as the output of the target recommendation model, and retraining the target recommendation model.

5. The method of claim 1, wherein the step of pushing each of the second programs to the designated device further comprises:

counting the target number of the second program;

if the number of the second programs is smaller than the number of the first multiple, or the number of the second programs is larger than the number of the second multiple, adjusting the similarity threshold, and counting the number of the second programs with the similarity larger than the adjusted similarity threshold again until the number of the second programs is larger than or equal to the number of the first multiple and smaller than or equal to the number of the second multiple, stopping adjusting the similarity threshold, and outputting the obtained second programs;

and pushing each second program to the designated equipment.

6. The method of claim 1, wherein the step of calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models and comparing the similarities comprises:

receiving search information input by a user, and analyzing a first dimension value corresponding to a search dimension of the search information;

comparing the search dimension value with a second dimension value of the first program vector corresponding to the search dimension;

judging whether the difference value between the first dimension value and the second dimension value is within a preset range or not;

if the program vector is within a preset range, giving a preset weight to the first program vector corresponding to the search dimension;

and calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models based on the preset weight, and comparing the similarity.

7. A program recommendation apparatus based on a target program recommendation model, comprising:

the extraction module is used for extracting a first program type and first program information of a first program currently played by the appointed equipment;

the vectorization module is used for vectorizing the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

the summing module is used for summing the first type vector and the first information vector according to a preset weighted summing mode to obtain a first program vector;

the second program information acquisition module is used for acquiring second program types and second program information of a plurality of programs played by a plurality of devices respectively;

the second program information vectorization module is used for vectorizing the second program type and the second program information to obtain a corresponding second type vector and a corresponding second information vector;

the second program vector solving module is used for summing the second type vector and the second information vector corresponding to each device according to a preset weighted summation mode to obtain a second program vector;

the clustering module is used for clustering each second program vector according to preset dimensions to obtain a training set corresponding to each preset dimension;

a combination module, configured to combine every two second program vectors in the training set in sequence and input the combined result into an input layer of an initial model corresponding to a preset dimension, where a formula v is adopted_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model;

a similarity value calculation module for calculating a similarity value according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein, Rkpq is a similarity value corresponding to the p-th second program vector and the q-th second program vector in the kth training set;

a discrimination module for discriminating in the discrimination network of the initial model according to a formulaCalculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

the adjusting module is used for adjusting parameters of the initial model based on the loss value corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimensionality after adjustment;

the calculation module is used for calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model and comparing the similarity;

the selecting module is used for selecting a target program recommending model with the highest similarity according to the comparison result, and inputting the first program vector into the target program recommending model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program;

and the pushing module is used for pushing each second program to the designated equipment.

8. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.

Technical Field

The invention relates to the field of digital televisions, in particular to a program recommendation method and device based on a target program recommendation model.

Background

With the application and development of network technology in the field of intelligent broadcast television in China, the digital television technology also brings new changes to the playing mode and the traditional broadcast technology of people. The network digital broadcasting and television technology has higher definition and more stable pictures; the television resources are richer; the playing surpasses the time and space limitation and has the advantage of low cost, thereby bringing excellent playing experience to television users. However, digital televisions are generally family-oriented, that is, facing a general community, and currently, existing intelligent recommendation of programs is generally recommended based on information or history records of clients, for example, the users like sports, and sports programs can be recommended to the users in 26 years old. However, for families, especially the elderly and children, the users generally do not set their own exclusive tags, and only one account is used, so that it is difficult to recommend a television program to a person in front of a digital television when the person is not determined.

Disclosure of Invention

The invention mainly aims to provide a program recommendation method, a device, equipment and a storage medium based on a target program recommendation model, aiming at solving the problem that when a person in front of a digital television is not determined, related television programs are difficult to recommend to the person.

The invention provides a program recommendation method based on a target program recommendation model, which comprises the following steps:

extracting a first program type and first program information of a first program currently played by a designated device;

vectorizing the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

summing the first type vector and the first information vector according to a preset weighted summing mode to obtain a first program vector;

acquiring second program types and second program information of a plurality of programs played by a plurality of devices respectively;

vectorizing the second program type and the second program information to obtain a corresponding second type vector and a corresponding second information vector;

summing the second type vector and the second information vector corresponding to each device according to a preset weighted summation mode to obtain a second program vector;

clustering each second program vector according to preset dimensions to obtain a training set corresponding to each preset dimension;

combining every two second program vectors in the training set in pairs and sequentially inputting the combined second program vectors into an input layer of an initial model corresponding to a preset dimension by adopting a formula v_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model;

according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein, Rkpq is a similarity value corresponding to the p-th second program vector and the q-th second program vector in the kth training set;

in the discrimination network of the initial model, according to a formula

Calculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

adjusting parameters of the initial model based on the loss values corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimension after adjustment;

calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model, and comparing the similarity;

selecting a target program recommendation model with the highest similarity according to the comparison result, and inputting the first program vector into the target program recommendation model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program;

and pushing each second program to the designated equipment.

Further, the step of extracting the first program information from the first program type and the first program information of the first program currently played by the specified device includes:

acquiring a search word of the designated device based on the first program, and determining a related word set of the search word, wherein the related word set comprises one or more related words;

determining one or more target related words from the related word set according to the first part of speech of the search word and the second part of speech of each related word;

determining a plurality of reference devices according to the search term and the one or more target related terms; wherein the reference device is a device that searches the first program using the search term and at least one of the one or more target related terms;

and determining common characteristics of the reference devices, and recording the common characteristics as the first program information.

Further, before the step of pushing each second program to the designated device, the method further includes:

acquiring target program vectors corresponding to a plurality of target programs of which the playing time of the appointed equipment is greater than a set value; wherein the target program vector comprises a program type of the target program and the program information;

extracting the dimension value of each dimension in each target program vector from preset dimensions;

extracting dimension values of all dimensions corresponding to all target program vectors according to all the preset dimensions respectively to obtain a dimension value set corresponding to all the preset dimensions respectively;

extracting the maximum value and the minimum value in each dimension set;

according to the formulaCalculating a standard value of a dimension value corresponding to each dimension set, wherein x is_ijRepresents the ith dimension value, min (x) in the jth dimension value data set_ij) Represents the minimum value, max (x), of the element in the jth of the dimension numerical data set_ij) Maximum value, Y, of an element in jth of the dimension numerical data set_ijRepresenting the standard value corresponding to the ith dimension value in the dimension value data set;

according to the formulaCalculating information entropy values for each of the sets of dimensions, whereinWherein E is_jSaid information entropy value representing the jth of said dimension set, when P_ijWhen =0, defineRepresenting a probability value corresponding to the ith dimension value of the jth dimension set, wherein n represents the number of the dimension sets;

comparing the information entropy value of each dimensionality set with preset information entropy values corresponding to each preset dimensionality;

screening out the dimension set smaller than the corresponding preset information entropy value according to a comparison result, and recording as a target dimension set;

recording the minimum value in each target dimension set as a dimension requirement value corresponding to a preset dimension;

obtaining a dimension value of each second program corresponding to the dimension of the target dimension set, and judging whether the dimension value is greater than the corresponding dimension requirement value;

and if the dimension requirement value is larger than the corresponding dimension requirement value, judging that the corresponding second program meets the requirement of pushing the second program to the specified equipment.

Further, after the step of pushing each second program to the designated device, the method further includes:

acquiring a second program selected by a user on the designated equipment, and recording the second program as a second target program;

extracting program information and playing integrity of each second target program;

converting the integrity degree into a corresponding recommended score according to a preset score conversion method;

and using the program information as the input of the target recommendation model, using the recommendation score as the output of the target recommendation model, and retraining the target recommendation model.

Further, the step of pushing each second program to the designated device further includes:

counting the target number of the second program;

if the number of the second programs is smaller than the number of the first multiple, or the number of the second programs is larger than the number of the second multiple, adjusting the similarity threshold, and counting the number of the second programs with the similarity larger than the adjusted similarity threshold again until the number of the second programs is larger than or equal to the number of the first multiple and smaller than or equal to the number of the second multiple, stopping adjusting the similarity threshold, and outputting the obtained second programs;

and pushing each second program to the designated equipment.

Further, the step of calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models and comparing the similarities includes:

receiving search information input by a user, and analyzing a first dimension value corresponding to a search dimension of the search information;

comparing the search dimension value with a second dimension value of the first program vector corresponding to the search dimension;

judging whether the difference value between the first dimension value and the second dimension value is within a preset range or not;

if the program vector is within a preset range, giving a preset weight to the first program vector corresponding to the search dimension;

and calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models based on the preset weight, and comparing the similarity.

The invention provides a program recommending device based on a target program recommending model, which comprises the following components:

the extraction module is used for extracting a first program type and first program information of a first program currently played by the appointed equipment;

the vectorization module is used for vectorizing the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

the summing module is used for summing the first type vector and the first information vector according to a preset weighted summing mode to obtain a first program vector;

the second program information acquisition module is used for acquiring second program types and second program information of a plurality of programs played by a plurality of devices respectively;

the second program information vectorization module is used for vectorizing the second program type and the second program information to obtain a corresponding second type vector and a corresponding second information vector;

the second program vector solving module is used for summing the second type vector and the second information vector corresponding to each device according to a preset weighted summation mode to obtain a second program vector;

the clustering module is used for clustering each second program vector according to preset dimensions to obtain a training set corresponding to each preset dimension;

a combination module, configured to combine every two second program vectors in the training set in sequence and input the combined result into an input layer of an initial model corresponding to a preset dimension, where a formula v is adopted_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model;

a similarity value calculation module for calculating a similarity value according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein, Rkpq is a similarity value corresponding to the p-th second program vector and the q-th second program vector in the kth training set;

a discrimination module for discriminating in the discrimination network of the initial model according to a formulaCalculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

the adjusting module is used for adjusting parameters of the initial model based on the loss value corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimensionality after adjustment;

the calculation module is used for calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model and comparing the similarity;

the selecting module is used for selecting a target program recommending model with the highest similarity according to the comparison result, and inputting the first program vector into the target program recommending model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program;

and the pushing module is used for pushing each second program to the designated equipment.

The invention also provides a computer device comprising a memory storing a computer program and a processor implementing the steps of any of the above methods when the processor executes the computer program.

The invention also provides a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method of any of the above.

The invention has the beneficial effects that: the method comprises the steps of obtaining a first program vector by extracting a first program type and first program information of a first program currently played by a designated device, calculating similarity between the first program vector and recommendation vectors preset by program recommendation models, comparing the similarity, selecting a target program recommendation model with the highest similarity according to a comparison result, inputting the first program vector into the target program recommendation model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program, and pushing the second programs to the designated device. Therefore, the program recommendation can be completed directly according to the first program currently played by the user without acquiring the user information of the user, the program recommendation accuracy is improved by acquiring the related recommendation model, and the problem that related television programs are difficult to recommend to people before the digital television is not determined is solved.

Drawings

Fig. 1 is a flowchart illustrating a program recommendation method based on a target program recommendation model according to an embodiment of the present invention;

fig. 2 is a schematic block diagram of a program recommendation apparatus based on a target program recommendation model according to an embodiment of the present invention;

fig. 3 is a block diagram illustrating a structure of a computer device according to an embodiment of the present application.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. 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.

It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship between the components, the motion situation, etc. in a specific posture (as shown in the drawings), and if the specific posture is changed, the directional indicator is changed accordingly, and the connection may be a direct connection or an indirect connection.

The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and B, may mean: a exists alone, A and B exist simultaneously, and B exists alone.

In addition, the descriptions related to "first", "second", etc. in the present invention are only for descriptive purposes and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In addition, technical solutions between various embodiments may be combined with each other, but must be realized by a person skilled in the art, and when the technical solutions are contradictory or cannot be realized, such a combination should not be considered to exist, and is not within the protection scope of the present invention.

Referring to fig. 1, the present invention provides a program recommendation method based on a target program recommendation model, including:

s1: extracting a first program type and first program information of a first program currently played by a designated device;

s2: vectorizing the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

s3: summing the first type vector and the first information vector according to a preset weighted summing mode to obtain a first program vector;

s4: acquiring second program types and second program information of a plurality of programs played by a plurality of devices respectively;

s5: vectorizing the second program type and the second program information to obtain a corresponding second type vector and a corresponding second information vector;

s6: summing the second type vector and the second information vector corresponding to each device according to a preset weighted summation mode to obtain a second program vector;

s7: clustering each second program vector according to preset dimensions to obtain a training set corresponding to each preset dimension;

s8: combining every two second program vectors in the training set in pairs and sequentially inputting the combined second program vectors into an input layer of an initial model corresponding to a preset dimension by adopting a formula v_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model; (ii) a

S9: according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein, Rkpq is a similarity value corresponding to the p-th second program vector and the q-th second program vector in the kth training set;

s10: in the discrimination network of the initial model, according to a formulaCalculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

s11: adjusting parameters of the initial model based on the loss values corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimension after adjustment;

s12: calculating the similarity between the first program vector and recommendation vectors preset by each program recommendation model, and comparing the similarity; wherein, different program recommendation models are trained by programs corresponding to the types of the programs;

s13: selecting a target program recommendation model with the highest similarity according to the comparison result, and inputting the first program vector into the target program recommendation model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program;

s14: and pushing each second program to the designated equipment.

As described in the above step S1, the first program type and the first program information of the first program currently played by the specified device are extracted. The designated device is a digital television in a home, processes signals in a digital mode, and generally comprises: digital high definition television (HDTV, movie-level images), digital enhanced definition television (EDTV, slightly higher images than DVD-level images), digital standard definition television (SDTV, DVD-level images), and digital popular television (VCD-level images). A first program type of a first program currently played by the specified device and first program information may be obtained at the specified device, where it is to be noted that the first program type may be obtained from a label that is pre-marked for the first program, for example, for a variety, sports, and the like, and the first program information includes a program name of the program, a program profile, and the like, which may embody characteristics of the first program.

As described in step S2, the first program type and the first program information are vectorized to obtain a corresponding first type vector and a corresponding first information vector. The vectorization mode may be implemented by a support vector machine SVM, the support vector machine is formed by training different program types or different program information and corresponding vectors in advance, it should be noted that the vector generated by the support vector machine is a vector composed of a plurality of preset dimensions, that is, the support vector machine may obtain dimension values of each dimension of the corresponding program type or program information, and then sort the vectors according to the sorting mode of the preset dimensions, so as to obtain the corresponding vectors.

As described in step S3, the first type vector and the first information vector are summed according to a preset weighted summation manner to obtain a first program vector. The preset weighted summation mode is to preset a first weight of the first type vector and a second weight of the first information vector. And then multiplying the first type vector and the first information vector by the corresponding first weight and the second weight respectively, and adding to obtain the corresponding first program vector, so that the first program vector has the information of the first type vector and the first information vector, and the information is integrated.

As described in the above steps S4-S11, training of recommendation models corresponding to respective programs is realized.

In step S4, the second program type and the second program information of the programs played by the respective devices are obtained. The obtaining mode may be obtained from data played by each device, and when each device plays a program, the device receives program data of a related program, so that the second program type and the second program information of a plurality of corresponding programs may be obtained from the server.

In step S5, the second program type and the second program information are vectorized to obtain a corresponding second type vector and a second information vector. The vectorization may be implemented by a support vector machine SVM, the support vector machine is formed by training different program types or different program information and corresponding vectors in advance, it should be noted that the vector generated by the support vector machine is a vector composed of a plurality of preset dimensions, that is, the support vector machine may obtain dimension values of each dimension of the corresponding program type or program information, and then sort the vectors according to the sorting mode of the preset dimensions, so as to obtain the corresponding vectors.

In step S6, the second type vector and the second information vector corresponding to each device are summed according to a preset weighted summation manner to obtain a second program vector. And summing the second type vector and the second information vector according to a preset weighted summation mode to obtain a second program vector. The preset weighted summation mode is to preset a first weight of the second type vector and a second weight of the second information vector. And then multiplying the second type vector and the second information vector by the corresponding first weight and the second weight respectively, and adding to obtain a corresponding second program vector, so that the second program vector has the information of the second type vector and the second information vector, and the information is integrated.

In step S7, each of the second program vectors is clustered according to a preset dimension to obtain a training set corresponding to each preset dimension, where the second program vectors and the preset dimension may be clustered by setting a corresponding cluster center point based on each dimension, calculating a distance from each second program vector to each cluster center point, and setting the preset dimension corresponding to the cluster center point where the distance between each second program vector is shortest as the dimension of each second program vector.

In step S8, since the similarity between the second program vectors is trained, the second program vectors need to be combined two by two to be used as a set of training data, and input into the input layer of the initial model. Using the formula v_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents an angular frequency of the characteristic function, j is a constant, and t is a period of the characteristic function, wherein the initial model is a generative confrontation network model.

In step S9, according to formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) to obtain a similarity value between the two. The larger the dot product value of the two is, the more the vectors representing the two are the same, so the dot product of the two is taken as a similarity value, and the similarity between the two can be well reflected.

In steps S10-S11, in the discriminant network of the initial model, the method is based on the formula

Calculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter, baseAnd adjusting parameters of the initial model according to the loss values corresponding to the initial model, and obtaining a program recommendation model corresponding to each dimension after adjustment. Wherein, in the formula, adoptAs a punishment item of the overfitting, the problem of overfitting of sparse data caused by summing the second program vector through the second type vector and the second information vector according to a preset weighted summation mode can be solved, and therefore an optimized loss value can be obtained. And training the confrontation network model through corresponding generation until the loss value reaches a preset value. Thereby completing the training of each of the different types of recommendation models.

As described in step S12, the similarity between the first program vector and the recommendation vectors preset by the program recommendation models is calculated, and similarity comparison is performed. Each program recommendation model is a preset program recommendation model, and in order to improve recommendation accuracy, different recommendation models are trained through different types of training data, for example, a sports-type program recommendation model is trained by using similarity between sports programs and program vectors corresponding to the sports programs, so that the corresponding recommendation vectors are sports-related vectors, and similarity comparison can be performed by calculating similarity between a first program vector and a recommendation vector preset by each program recommendation model. Therefore, an ideal program recommendation model can be selected from the program recommendation models, and the program recommendation accuracy is improved.

As described in step S13, the target program recommendation model with the highest similarity is selected according to the comparison result, and the first program vector is input into the target program recommendation model, so as to obtain a plurality of second programs with similarity greater than the similarity threshold value with the first program. Namely, the first program vector is input and a program recommendation model is used for obtaining a plurality of corresponding second programs with the similarity greater than the similarity threshold value with the first program. It should be noted that the program vector of the second program is stored in the program recommendation model in advance, and the corresponding second program can be obtained only by inputting the first program vector.

As described in the above step S14, each of the second programs is pushed to the designated device. The second programs are pushed to the designated equipment, so that the program recommendation can be completed directly according to the first program currently played by the user without acquiring the user information of the user, the program recommendation accuracy is improved by acquiring the related recommendation model, and the problem that the related television programs are difficult to recommend to the personnel before the digital television is not determined is solved.

In one embodiment, the step S1 of extracting the first program information from the first program information and the first program type of the first program currently played by the specified device includes:

s101: acquiring a search word of the designated device based on the first program, and determining a related word set of the search word, wherein the related word set comprises one or more related words;

s102: determining one or more target related words from the related word set according to the first part of speech of the search word and the second part of speech of each related word;

s103: determining a plurality of reference devices according to the search term and the one or more target related terms; wherein the reference device is a device that searches the first program using the search term and at least one of the one or more target related terms;

s104: and determining common characteristics of the reference devices, and recording the common characteristics as the first program information.

As described in the above steps S101 to S104, the first program information is obtained, where it should be noted that the first program information may also include features of each device playing the first program.

In steps S101-S103, in the present embodiment, the related word is a vocabulary linked to the preset knowledge base, and includes one of the multiple meaning words of the search word, the search word itself, and the synonym of the search word. The terminal equipment determines related words of the search words, including determining whether the search words are polysemous words or not, if the search words are polysemous words, determining the meanings of the search words, and accordingly determining that the knowledge base links are words corresponding to the meanings of the search words or synonyms corresponding to the words, and if the search words are not polysemous words, determining that the related words of the search words are the search words or the synonyms of the search words. Alternatively, whether the search word is a polysemous word may be determined according to a preset dictionary, and the preset dictionary includes a plurality of ambiguous words of the search word, which indicates that the search word is a polysemous word. When the search term is not a polysemous term, the related term is the search term itself or a synonym of the search term, for example, the search term is a support vector machine, and the related term can be a support vector machine, a support vector network, and an SVM. And when the search word is a multi-meaning word, determining the meaning of the search word in the text content to determine the related word of the search word. So that the reference device which has searched using the related word or the search word can be determined.

In step S104, a common feature of each of the reference devices is determined, and the common feature is recorded as the first program information. Since each reference device can be regarded as having a common point with the user currently playing the first program, it can be regarded as having the same preference, and thus can be used as program information of the first program to perform the similarity calculation basis of the second program. Therefore, the technical effect that the user information of the user does not need to be obtained in advance and the second program can be recommended to the user can be achieved.

In an embodiment, before the step S14 of pushing each of the second programs to the designated device, the method further includes:

s1301: acquiring target program vectors corresponding to a plurality of target programs of which the playing time of the appointed equipment is greater than a set value; wherein the target program vector comprises a program type of the target program and the program information;

s1302: extracting the dimension value of each dimension in each target program vector from preset dimensions;

s1303: extracting dimension values of all dimensions corresponding to all target program vectors according to all the preset dimensions respectively to obtain a dimension value set corresponding to all the preset dimensions respectively;

s1304: extracting the maximum value and the minimum value in each dimension set;

s1305: according to the formulaCalculating a standard value of a dimension value corresponding to each dimension set, wherein x is_ijRepresents the ith dimension value, min (x) in the jth dimension value data set_ij) Represents the minimum value, max (x), of the element in the jth of the dimension numerical data set_ij) Maximum value, Y, of an element in jth of the dimension numerical data set_ijRepresenting the standard value corresponding to the ith dimension value in the dimension value data set;

s1306: according to the formulaCalculating information entropy values for each of the sets of dimensions, whereinWherein E is_jSaid information entropy value representing the jth of said dimension set, when P_ijWhen =0, defineRepresenting a probability value corresponding to the ith dimension value of the jth dimension set, wherein n represents the number of the dimension sets;

s1307: comparing the information entropy value of each dimensionality set with preset information entropy values corresponding to each preset dimensionality;

s1308: screening out the dimension set smaller than the corresponding preset information entropy value according to a comparison result, and recording as a target dimension set;

s1309: recording the minimum value in each target dimension set as a dimension requirement value corresponding to a preset dimension;

s1310: obtaining a dimension value of each second program corresponding to the dimension of the target dimension set, and judging whether the dimension value is greater than the corresponding dimension requirement value;

s1311: and if the dimension requirement value is larger than the corresponding dimension requirement value, judging that the corresponding second program meets the requirement of pushing the second program to the specified equipment.

As described in step S1301 above, a plurality of target programs whose playing time is greater than the set value are acquired according to the designated device. Namely, the appointed equipment counts the time of the played program, so that the target program on the appointed equipment can be directly obtained, and then vectorization is carried out through a preset vector machine, so that the corresponding target program vector can be obtained.

As described in step S1302, the dimension information of the target program vector is extracted from the preset dimensions, and the dimension information is converted into a dimension value according to the preset corresponding relationship. The preset dimension is a dimension set in advance, the preset dimension can be generated according to a designated device, different dimension requirements (the dimension requirement is a requirement in each aspect of the recruitment information) exist in the designated device, the corresponding relation between the dimension information and the dimension value can be set in advance, and the corresponding dimension value can be obtained according to the dimension information subsequently.

As described in the above steps S1303 to S1304, the dimension values corresponding to the target program vectors are extracted according to each dimension, so as to obtain a dimension value set corresponding to each dimension; namely, the same dimensionality in each resume vector is extracted, and a dimensionality set is constructed for each dimensionality. And then extracting the maximum value and the minimum value in each dimension set according to the size of each dimension value.

As described in step S1305, information entropy of each dimension value set is calculated, where the formula of the calculation may be to calculate a variance or an average difference in the dimension value set, that is, a maximum value and a minimum value in the dimension value set are obtained first, and data fluctuation of the entire dimension value set is reflected according to the maximum value and the minimum value, that is, the data fluctuation is reflected according to the formula firstCalculating the standard value corresponding to each dimension value, namely firstly carrying out standard treatment on each dimension valueAnd the normalization processing is carried out, so that the deviation of the calculation result caused by overlarge data is avoided. Then, according to the probability P of the standard deviation corresponding to each dimension value_ijAnd calculating the information entropy value of the jth dimension value set. The information entropy value calculated according to the calculation formula fully considers the fluctuation condition of each dimensionality value in the same dimensionality and also fully considers the influence of the extreme individual value on the whole information entropy value, so that the calculated information entropy value has higher referential property.

As described in the above steps S1307-S1308, the dimension requirement of the corresponding dimension is set according to the corresponding information entropy calculated by each dimension value set, and it should be understood that when the information entropy of a certain dimension value set is smaller, it indicates that the corresponding dimension value is relatively regular, and indicates that the dimension has a preference for the program of the corresponding dimension, so that the dimension set smaller than the corresponding preset information entropy is screened out according to the comparison result and is recorded as the target dimension set.

As described in the above steps S1309-S1311, the minimum value in each target dimension set is recorded as the dimension requirement value corresponding to the preset dimension, the dimension value of each second program corresponding to the target dimension set dimension is obtained, and it is determined whether the value is greater than the corresponding dimension requirement value. And if the dimension requirement value is larger than the corresponding dimension requirement value, judging that the corresponding second program meets the requirement of pushing the second program to the specified equipment. Therefore, the screening of the preset dimensionality is realized, the preference of the user is determined, and the second program which is larger than the dimensionality required value is pushed. Thereby improving the user's satisfaction with the pushed second program.

In an embodiment, after the step S14 of pushing each of the second programs to the designated device, the method further includes:

s1501: acquiring a second program selected by a user on the designated equipment, and recording the second program as a second target program;

s1502: extracting program information and playing integrity of each second target program;

s1503: converting the integrity degree into a corresponding recommended score according to a preset score conversion method;

s1504: and using the program information as the input of the target recommendation model, using the recommendation score as the output of the target recommendation model, and retraining the target recommendation model.

As described above in steps S1501-S1504, retraining of the target recommendation model is achieved. Namely, the second program selected by the user on the designated device is obtained and recorded as the second target program. And extracting the playing integrity of the second target program played by the user, if the playing integrity is very small, the user is not satisfied with the second target program, so that the corresponding recommendation score is small, and if the playing integrity is very large, the user is satisfied with the second target program, so that the corresponding recommendation score is large, then the program information is used as the input of the target recommendation model, and the recommendation score is used as the output of the target recommendation model, and the target recommendation model is retrained, so that the fitting degree of the recommendation model and a family can be continuously improved.

In an embodiment, the step S14 of pushing each of the second programs to the designated device further includes:

s1511: counting the target number of the second program;

s1512: if the number of the second programs is smaller than the number of the first multiple, or the number of the second programs is larger than the number of the second multiple, adjusting the similarity threshold, and counting the number of the second programs with the similarity larger than the adjusted similarity threshold again until the number of the second programs is larger than or equal to the number of the first multiple and smaller than or equal to the number of the second multiple, stopping adjusting the similarity threshold, and outputting the obtained second programs;

s1513: and pushing each second program to the designated equipment.

As described in the above steps S1511 to S1513, the limitation of the second program number is realized. If the number of the second programs is smaller than the number of the first multiple, or the number of the second programs is larger than the number of the second multiple, the similarity threshold value is adjusted until the number of the second programs is larger than or equal to the number of the first multiple and smaller than or equal to the number of the second multiple, the adjustment of the similarity threshold value is stopped, and the obtained second programs are output.

When the number of the second programs is smaller than the number of the first multiple, it means that the screened second programs are too few, the second programs related to the first program are few, and the similarity threshold needs to be adjusted to reduce the similarity requirement, improve the number of the screened second programs, and avoid that the proper second programs cannot be searched.

When the number of the second programs is larger than the second multiple, the screened second programs are too many, and the similarity threshold value needs to be adjusted to improve the similarity requirement, reduce the number of the screened second programs, and simultaneously reduce the burden of a user for browsing and screening each second program, so that the user can conveniently find the appropriate second program.

In an embodiment, the step S12 of calculating the similarity between the first program vector and the recommendation vectors preset by the respective program recommendation models, and performing similarity comparison includes:

s1201: receiving search information input by a user, and analyzing a first dimension value corresponding to a search dimension of the search information;

s1202: comparing the search dimension value with a second dimension value of the first program vector corresponding to the search dimension;

s1203: judging whether the difference value between the first dimension value and the second dimension value is within a preset range or not;

s1204: if the program vector is within a preset range, giving a preset weight to the first program vector corresponding to the search dimension;

s1205: and calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models based on the preset weight, and comparing the similarity.

As described in the above steps S1201-S1205, accurate query recommendation is realized in combination with the search content of the user. Specifically, the user may input related search information, for example, a basketball, and may analyze a dimension corresponding to the search information, that is, a sports dimension, and assuming that a dimension value corresponding to the basketball is 10, the search dimension value may be compared with a second dimension value of the first program vector corresponding to the search dimension. If the program currently watched by the user is a program related to basketball or slightly different from basketball, the difference value between the corresponding first dimension value and the second dimension value is quite small, the type of the first program is shown to be watched, the type of the second program to be recommended is the same, the recommendation can be continued, the first program vector is endowed with the preset weight corresponding to the search dimension, namely, the weight is higher, the occupation ratio of the corresponding dimension of the search content of the user is improved, and the obtained second program can better meet the user.

Referring to fig. 2, the present invention further provides a program recommendation apparatus based on a target program recommendation model, including:

the extracting module 10 is configured to extract a first program type and first program information of a first program currently played by a designated device;

a vectorization module 20, configured to vectorize the first program type and the first program information to obtain a corresponding first type vector and a corresponding first information vector;

the summing module 30 is configured to sum the first type vector and the first information vector according to a preset weighted summing manner to obtain a first program vector;

a second program information obtaining module 40, configured to obtain second program types and second program information of multiple programs played by multiple devices respectively;

a second program information vectorization module 50, configured to vectorize the second program type and the second program information to obtain a corresponding second type vector and a second information vector;

a second program vector calculation module 60, configured to sum the second type vector and the second information vector corresponding to each device according to a preset weighted sum manner, so as to obtain a second program vector;

a clustering module 70, configured to perform clustering processing on each second program vector according to a preset dimension to obtain a training set corresponding to each preset dimension;

a combining module 80, configured to combine every two second program vectors in the training set in sequence and input the combined result into an input layer of the initial model corresponding to the preset dimension, where a formula v is used_ik=w_j×f（x_ik) + b, obtaining the characteristic vectors corresponding to the second program vectors respectively; wherein v is_ikRepresenting the feature vector, x, corresponding to the ith second program vector in the kth training set_ikRepresenting the ith second program vector, w, in the kth training set_jThe preset total weight of training for the jth dimension, b is a preset bias parameter, f (x)_ik)=exp(jwt)*x_ikExp (jwt) is a characteristic function of the model, w represents the angular frequency of the characteristic function, j is a constant, t is the period of the characteristic function, wherein the initial model is a generation countermeasure network model;

a similarity value calculation module 90 for calculating a similarity value according to the formula R_kpq=v_pk ^Tv_qk=(w_j×f(v_pk)+b)^T(w_j×f(v_qk) + b) obtaining the similarity value of the two; wherein, Rkpq is a similarity value corresponding to the p-th second program vector and the q-th second program vector in the kth training set;

a discriminant module 100 for discriminating the network of the initial model according to a formulaCalculating a loss value of each initial model based on the training set; where n denotes the number of second program vectors in the kth training set, r_kpqRepresenting a feature vector v_qkAnd the feature vector v_pkThe actual value of similarity between the two,is a preset parameter;

and the adjusting module 110 is configured to perform parameter adjustment on the initial model based on the loss value corresponding to the initial model, and obtain a program recommendation model corresponding to each dimension after the parameter adjustment.

The calculating module 120 is configured to calculate similarity between the first program vector and recommendation vectors preset by each program recommendation model, and perform similarity comparison;

the selecting module 130 is configured to select a target program recommendation model with the highest similarity according to the comparison result, and input the first program vector into the target program recommendation model to obtain a plurality of second programs with similarity greater than a similarity threshold value with the first program;

the pushing module 1400 is configured to push each of the second programs to the designated device.

In one embodiment, the program recommending apparatus based on the target program recommendation model further includes:

in one embodiment, the extraction module 10 includes:

a search word obtaining sub-module, configured to obtain a search word of the first program based on the designated device, and determine a related word set of the search word, where the related word set includes one or more related words;

the target related word determining submodule is used for determining one or more target related words from the related word set according to the first part of speech of the search word and the second part of speech of each related word;

a reference device determining sub-module, configured to determine multiple reference devices according to the search term and the one or more target related terms; wherein the reference device is a device that searches the first program using the search term and at least one of the one or more target related terms;

and the common characteristic determining submodule is used for determining the common characteristic of each reference device and recording the common characteristic as the first program information.

In one embodiment, the program recommending apparatus based on the target program recommendation model further includes:

the target program vector acquisition module is used for acquiring target program vectors corresponding to a plurality of target programs of which the playing time of the appointed equipment is greater than a set value; wherein the target program vector comprises a program type of the target program and the program information;

the dimension value extraction module is used for extracting the dimension value of each dimension in each target program vector from preset dimensions;

the dimension value set acquisition module is used for respectively extracting the dimension values of all dimensions corresponding to all target program vectors according to all the preset dimensions to obtain a dimension value set corresponding to all the preset dimensions;

the maximum value extraction module is used for extracting the maximum value and the minimum value in each dimension set;

a standard value calculation module for calculating the standard value according to a formulaCalculating a standard value of a dimension value corresponding to each dimension set, wherein x is_ijRepresents the ith dimension value, min (x) in the jth dimension value data set_ij) Represents the minimum value, max (x), of the element in the jth of the dimension numerical data set_ij) Maximum value, Y, of an element in jth of the dimension numerical data set_ijRepresenting the standard value corresponding to the ith dimension value in the dimension value data set;

an information entropy calculation module for calculating the entropy according to a formulaCalculating information entropy values for each of the sets of dimensions, whereinWherein E is_jSaid information entropy value representing the jth of said dimension set, when P_ijWhen =0, defineRepresenting the probability corresponding to the ith dimension value of the jth dimension setA value, n, representing the number of said set of dimensions;

the information entropy value comparison module is used for comparing the information entropy value of each dimensionality set with a preset information entropy value corresponding to each preset dimensionality;

the dimension set screening module is used for screening out the dimension set smaller than the corresponding preset information entropy value according to a comparison result and recording the dimension set as a target dimension set;

the dimension requirement value setting module is used for recording the minimum value in each target dimension set as a dimension requirement value corresponding to a preset dimension;

the dimension value judging module is used for acquiring the dimension value of each second program corresponding to the dimension of the target dimension set and judging whether the dimension value is larger than the corresponding dimension requirement value;

and the second program judging module is used for judging that the corresponding second program meets the requirement of pushing the second program to the specified equipment if the second program is larger than the corresponding dimension requirement value.

In one embodiment, the program recommending apparatus based on the target program recommendation model further includes:

acquiring a second program selected by a user on the designated equipment, and recording the second program as a second target program;

the integrity extraction module is used for extracting the program information and the playing integrity of each second target program;

the conversion module is used for converting the integrity degree into a corresponding recommendation score according to a preset score conversion method;

and the retraining module is used for retraining the target recommendation model by taking the program information as the input of the target recommendation model and taking the recommendation score as the output of the target recommendation model.

In one embodiment, the push module 140 includes:

the target quantity counting submodule is used for counting the target quantity of the second program;

a number adjusting submodule, configured to adjust the similarity threshold if the number of the second programs is smaller than a first multiple number, or the number of the second programs is greater than a second multiple number of people, and count the number of the second programs whose similarity is greater than the adjusted similarity threshold again until the number of the second programs is greater than or equal to the first multiple number and is less than or equal to the second multiple number, stop adjusting the similarity threshold, and output the obtained second program;

and the second program pushing submodule is used for pushing each second program to the designated equipment.

In one embodiment, the calculation module 120 includes:

the search information receiving submodule is used for receiving search information input by a user and analyzing a first dimension value corresponding to a search dimension of the search information;

a first program vector comparison submodule for comparing the search dimension value with a second dimension value of the first program vector corresponding to the search dimension;

a difference value judgment submodule for judging whether the difference value between the first dimension value and the second dimension value is within a preset range;

the giving sub-module is used for giving the first program vector a preset weight corresponding to the search dimension if the first program vector is within a preset range;

and the similarity calculation operator module is used for calculating the similarity between the first program vector and the recommendation vectors preset by the program recommendation models based on the preset weight and comparing the similarity.

The invention has the beneficial effects that: the method comprises the steps of obtaining a first program vector by extracting a first program type and first program information of a first program currently played by a designated device, calculating similarity between the first program vector and recommendation vectors preset by program recommendation models, comparing the similarity, selecting a target program recommendation model with the highest similarity according to a comparison result, inputting the first program vector into the target program recommendation model to obtain a plurality of second programs with the similarity larger than a similarity threshold value with the first program, and pushing the second programs to the designated device. Therefore, the program recommendation can be completed directly according to the first program currently played by the user without acquiring the user information of the user, the program recommendation accuracy is improved by acquiring the related recommendation model, and the problem that related television programs are difficult to recommend to people before the digital television is not determined is solved.

Referring to fig. 3, a computer device, which may be a server and whose internal structure may be as shown in fig. 3, is also provided in the embodiment of the present application. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the computer designed processor is used to provide computational and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for storing various program types, program information and the like. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, may implement the program recommendation method based on the target program recommendation model according to any of the embodiments described above.

Those skilled in the art will appreciate that the architecture shown in fig. 3 is only a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects may be applied.

The embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method for recommending a program based on a target program recommendation model according to any of the embodiments above may be implemented.

It will be understood by those skilled in the art that all or part of the processes of the methods of the above embodiments may be implemented by hardware associated with instructions of a computer program, which may be stored on a non-volatile computer-readable storage medium, and when executed, may include processes of the above embodiments of the methods. Any reference to memory, storage, database, or other medium provided herein and used in the examples may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and bus dynamic RAM (RDRAM).

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.