阅读说明：本技术 消费数据处理方法与系统 (Consumption data processing method and system ) 是由 李荣生 简嘉宏 傅心玮 陈钰欣 于 2020-05-22 设计创作，主要内容包括：一种消费数据处理方法与系统,该方法运行于一伺服系统中。一开始,伺服系统输入一企业数据,企业数据涵盖属于一企业属性的消费者数据,再对企业数据进行数据清洗,并获取其中特征,接着执行一机器学习法,取得企业数据中的卷标,建立一企业特性模型,再输入一主数据至此企业特性模型,以原先企业数据的少数特征,从主数据中建立一营销清单,之后可以依据此营销清单投放营销广告,并根据营销结果取得更多特征,建立更多卷标,以更新此企业特性模型。以此可优化关联企业属性的企业特性模型并持续更新优化。(A consumption data processing method and system are provided, the method is operated in a servo system. Firstly, a servo system inputs enterprise data, the enterprise data covers consumer data belonging to an enterprise attribute, then the enterprise data is subjected to data cleaning, characteristics of the enterprise data are obtained, then a machine learning method is executed, volume labels in the enterprise data are obtained, an enterprise characteristic model is established, main data are input into the enterprise characteristic model, a marketing list is established from the main data according to a few characteristics of the original enterprise data, then marketing advertisements can be launched according to the marketing list, more characteristics are obtained according to marketing results, and more volume labels are established to update the enterprise characteristic model. Thereby optimizing the enterprise personality model associated with the enterprise attributes and continually updating the optimization.)

1. A consumption data processing method, operating on a server system, the method comprising:

inputting enterprise data, wherein the enterprise data covers consumer data belonging to an enterprise attribute;

carrying out data cleaning on the enterprise data and acquiring characteristics of the enterprise data;

executing a machine learning method on the data which is subjected to data cleaning and characteristic acquisition to acquire the volume label in the enterprise data and establish an enterprise characteristic model;

inputting a master data into the enterprise personality model, wherein the master data covers a larger number of consumers than the enterprise data and is not limited to attributes of the enterprise data;

obtaining a marketing list of the characteristics related to the enterprise data from the main data;

putting marketing advertisements to all or part of consumers in the marketing list according to the marketing list; and

and updating the enterprise characteristic model according to a marketing result.

2. The consumption data processing method of claim 1, wherein in the step of performing data cleansing on the enterprise data, a customer relationship management analysis is performed to classify customers in the enterprise data.

3. The consumption data processing method of claim 2, wherein the customer relationship management analysis employs a last consumption, consumption frequency and consumption amount analysis to score the consumer in the enterprise data so as to obtain the volume label in the enterprise data.

4. The consumption data processing method of claim 2, wherein the step of performing data cleansing on the enterprise data comprises establishing a profile of each consumer in the enterprise data.

5. The consumption data processing method of claim 1, wherein the enterprise personality model associated with the enterprise attributes is optimized by iteratively importing the marketing results into the master data, retraining and updating the enterprise personality model with the machine learning method.

6. The consumption data processing method of any of claims 1 to 5, wherein the method of creating the master data comprises:

establishing one or more group buying activities in a community media;

analyzing the user's activities in the social media among the one or more group purchase activities; and

and establishing the main data according to the analysis result.

7. A consumption data processing system, characterized in that the system comprises:

a server system having a master database, in which a consumption data processing method is executed, comprising:

inputting enterprise data, wherein the enterprise data covers consumer data belonging to an enterprise attribute;

carrying out data cleaning on the enterprise data and acquiring characteristics of the enterprise data;

executing a machine learning method on the data which is subjected to data cleaning and characteristic acquisition to acquire the volume label in the enterprise data and establish an enterprise characteristic model;

inputting a master data into the enterprise personality model, wherein the master data covers a larger number of consumers than the enterprise data and is not limited to attributes of the enterprise data;

obtaining a marketing list of the characteristics related to the enterprise data from the main data;

putting marketing advertisements to all or part of consumers in the marketing list according to the marketing list; and

and updating the enterprise characteristic model according to a marketing result.

8. The consumption data processing system of claim 7, wherein in the step of performing data cleansing on the enterprise data in the consumption data processing method, a customer relationship management analysis is performed to classify customers in the enterprise data.

9. The consumption data processing system of claim 8, wherein the customer relationship management analysis employs a last consumption, consumption frequency, and consumption amount analysis to score the consumer within the enterprise data to facilitate obtaining the tags in the enterprise data.

10. The consumption data processing system of claim 7, wherein the enterprise personality model associated with the enterprise attributes is optimized by iteratively importing the marketing results into the master data, retraining the machine learning method, and updating the enterprise personality model.

11. The system as claimed in any one of claims 7 to 10, wherein the server system has a data platform connected to a community media host for obtaining consumption data formed in a community media through an api to establish the master data.

12. The consumption data processing system of claim 11, wherein the method for the server system to establish the master data via the social media comprises:

establishing one or more group purchase activities in the community media;

analyzing the user's activities in the social media among the one or more group purchase activities; and

and establishing the main data according to the analysis result.

Technical Field

The specification discloses a data processing method, in particular to a consumption data processing method and a system for summarizing existing consumption data, making up for data deviation and deficiency by using machine learning and achieving the marketing purpose.

Background

In the existing marketing mode, an advertiser (a business owner) puts demands on the advertiser, and after a text or a video media is made, the advertiser puts advertisements on public media, networks and social media (social medias).

The accurate marketing is that a target customer group (target audio) can be obtained according to the property of the commodity of the business, then the advertisement is delivered to the locked target customer group through a specific medium (mainly a network community), and the commodity (or service) of the delivered advertisement is consistent with the favor of the target customer group, so that the income benefit brought by the advertisement is relatively good.

The target customer base can be derived from data in the enterprise, and the consumers in the data can be the people who have come from the store (or network) for consumption in the past, and the traditional enterprise is not easy to extend to consumers beyond the data. The target customer group can also analyze objects screened from a plurality of consumer data through a big data analysis, for example, when advertisements are put in various social media, the target of putting the advertisements is set through selecting regions, age groups and the like, so that objects except the data of enterprises can be obtained, or a group of people with relative interest can be obtained through some screening mechanisms, and the put advertisements can be more accurate.

Disclosure of Invention

In view of the fact that general enterprises collect less consumption data and have deviation, the consumption data processing method and system provided by the specification utilize the data of the enterprises to learn the consumer characteristics of the enterprises through a machine learning algorithm so as to establish an enterprise characteristic model, and a group of new lists corresponding to the consumer characteristics of the enterprises are obtained through main data. In the consumption data processing method and system, one objective is to provide a method for enabling an enterprise to obtain more potential target customers besides own enterprise data, and when an enterprise characteristic model is established by adopting a machine learning method, a marketing list is provided according to the relevance of main data and enterprise data after the main data of a larger consumption record is input.

According to the embodiment, the provided consumption data processing method runs in a servo system, in the method, enterprise data is input, the enterprise data covers consumer data belonging to an enterprise attribute, then data cleaning is carried out on the enterprise data, characteristics of the enterprise data are obtained, a machine learning method is executed on the data subjected to the data cleaning and the characteristics obtaining, volume labels in the enterprise data are obtained, and an enterprise characteristic model is established. A master data may then be entered into the enterprise personality model, wherein the master data covers a greater number of consumers than the enterprise data and is not limited to attributes of the enterprise data.

Then, a marketing list of the characteristics of the associated enterprise data is obtained from the main data, marketing advertisements are put on all or part of the consumers according to the marketing list, the enterprise characteristic model can be updated according to a marketing result, and the enterprise characteristic model of the associated enterprise attributes can be optimized by repeatedly inputting and retraining the result to update the enterprise characteristic model.

Further, in the step of performing data cleansing on the enterprise data, a Customer Relationship Management (Customer Relationship Management) analysis may be performed to classify the customers in the enterprise data. The customer relationship management analysis may use a last time consumption (Recency), Frequency of consumption (Frequency), and amount of consumption (money) analysis to score customers within the enterprise data to obtain the tags in the enterprise data.

According to one embodiment, the master data is created by creating one or more group buying activities in a social media, then analyzing the information in the group buying activities, and then creating the master data according to the analysis result.

In an embodiment of the consumption data processing system, the system proposes a servo system provided with a master database in which the consumption data processing method is run.

Furthermore, the servo system is provided with a data platform for connecting the community media host, and consumption data formed in the community media can be obtained through an Application programming interface (Application Program Internet) to establish main data.

Drawings

FIG. 1 shows a schematic diagram of an embodiment of a system architecture for performing a method of consuming data processing;

FIG. 2 shows a schematic diagram of an embodiment of a consumer data processing system architecture;

FIG. 3 is a diagram of a flow embodiment of a method of consumption data processing;

FIG. 4 depicts a diagram of a method flow embodiment of creating master data;

FIG. 5 is a diagram illustrating a flow chart of a method for processing consumption data; and

fig. 6 shows a diagram of an embodiment of a method flow for verifying a marketing list.

Detailed Description

In view of the fact that general enterprises can only perform relatively accurate marketing promotion from past consumption data to increase the customer return rate on one hand and cannot effectively expand more consumer customer groups on the other hand, in the prior art, if more achievements are to be expanded, advertisers need to put advertisements to unspecific consumers by using various mass media, networks and community networks or put advertisements to simply screened target customer groups, but the advertisement effectiveness is not high because target customers cannot be effectively locked. Aiming at the problem that the data of general enterprises is few and has deviation, the consumption data processing method and the consumption data processing system are provided for establishing an enterprise characteristic model after learning the consumer characteristics of the enterprises by a machine learning algorithm so as to find a group of new lists which are consistent with the consumer characteristics of the enterprises from the provided main data, namely the problems of data deviation and data insufficiency are made up by an execution method, so that the enterprises can reach target customer groups which cannot be covered by the traditional data.

FIG. 1 is a schematic diagram of an embodiment of a system architecture for performing a method for consuming data.

In this example, the system side provides a server system 11, in which a main database 110 is provided, the system side is connected to the community medium 13 through the network 10, so as to obtain consumer data by consuming activities in the community medium 13, and on the other hand, the system side is connected to the enterprise host 15 through the network 10 so as to obtain data in the enterprise database 150, and after machine learning, an enterprise characteristic model is established, so that more clients can be reached through the main data of the system side, and deviations and deficiencies of data generated by enterprise attributes can be compensated.

On the other hand, the server 11 can establish its own sales system in the community medium 13, so that the supplier 17 of the goods or services can sell the goods or services through this sales channel, and at the same time, the server 11 can obtain a larger amount of consumer data through this sales channel, and establish master data, so as to expand business with the enterprise with small service data volume. After the marketing system is established by the machine learning method for various enterprise attributes, the system can provide the advertiser 19 to develop advertising business, and deliver marketing advertisements for marketing objects obtained by the enterprise characteristic model.

The concept of the consumption data processing method is that general enterprise data has deviation caused by enterprise attributes, the characteristics of enterprises are learned by a machine learning method to obtain the enterprise attributes, a plurality of characteristic labels can be used for describing the enterprise attributes, and then the system utilizes a larger amount of main data to obtain more potential target customer groups with similar attributes (having the same characteristic labels) according to a model formed by machine learning, so that the system can be a reference for new product development or shop exhibition suggestion besides new customer development.

Fig. 2 next depicts in flow chart a consumption data processing system architecture, wherein the enterprise side is schematically shown provided with an enterprise host 21, provided with an enterprise database 210, through which an enterprise data platform 212 may access clients 215. Enterprise data platform 212 is illustratively representative of consumer data obtained from various sources (e.g., a network) in addition to obtaining customer data in a conventional manner, wherein the enterprise data is consumer data corresponding to enterprise attributes.

The system side is provided with a servo system 20 and an enterprise data interface 201, which schematically shows that the enterprise data of the enterprise database 210 is obtained through the enterprise data interface 201, and a plurality of enterprises can be simultaneously served to obtain other enterprise data 208.

The server system 20 includes various functional modules implemented by software and hardware (such as a processor, a memory, and a network), as shown in the figure, a preprocessing module 202, a machine learning module 203, a model building module 204, a marketing list generating module 205, a main database 206 (including a member database and consumption data), and a data platform 207. Wherein each functional module can still be implemented by each independently operating host.

The preprocessing module 202 is used for performing data cleaning (data cleaning), feature acquisition, analysis, classification and other processing procedures on data input into the system, such as enterprise data. The data cleaning is to delete useless, incomplete, unrecognizable or repeated data from the input data, correct errors as much as possible, and then make various data uniformly accepted by the system through formatting. The program for acquiring the characteristics analyzes the content of each data for the meaning of the text to acquire the characteristics, and the characteristics acquired by the program for acquiring the characteristics aiming at the consumption data in this example comprise consumption areas, consumer ages, commodity prices, attributes, consumer occupation, income, consumption time, enterprise attributes and the like, and the characteristics are also expected to become volume labels required by establishing the model. The purpose of analysis and classification is to set parameters according to requirements to score each input data, for example, there is a common RFM scoring method, and objective indicators according to which include recent consumption (Recency), consumption Frequency (Frequency) and consumption amount (money) are used to classify each data, so as to obtain a tag in the enterprise data. It is mentioned here that RFM is only one of the feature extraction methods for the method, and is for illustration and not for limiting the proposed solution.

The machine learning module 203 is used to execute a specific machine learning algorithm to learn rules in the input data and obtain feature tags (tag) related to data attributes. In this example, training is performed according to big data in the enterprise database 210, data mining is performed, meaningful features (such as keywords) in various consumption data are obtained, a volume label associated with the main data is formed, and an enterprise characteristic model of the marketing object is obtained through the model building module 204.

The model building module 204 not only builds the enterprise characteristic model according to the feature tags of the enterprise data obtained by machine learning execution, but also updates the model according to feedback information generated by the marketing result, thereby achieving the purpose of optimizing the enterprise characteristic model.

Then, a marketing list is obtained from the main data through the marketing list generation module 205, the marketing list records suggested objects for later advertisement delivery, the suggested objects include a small amount of delivered marketing lists for testing, then the marketing result also becomes a basis for evaluating the goodness and badness of the enterprise characteristic model, and the marketing list generation module 205 continues to generate a new marketing list from the optimized enterprise characteristic model. Under a technical concept, firstly, a test result is put in for the first time, based on a few characteristics of original enterprise data, more various characteristics are obtained from main data with more orientations, so that a machine learning algorithm turns to obtain associated consumer characteristics from the main data, a marketing model with more dimensions is established, and then the optimized marketing model is continuously updated according to the next putting.

The server system 20 is provided with a main database 206, in which various consumer data are recorded, including member database, consumption data and the like obtained on various consumption platforms, and particularly including data formed by consumption activities in various community media 23, 24 obtained through a data platform 207, the data platform 207 is implemented in a customer relationship management and purchase-sale-storage system, and can reach users 250 in the community media 23, 24, and obtain data of consumption activities of the users 250 through Application Programming Interfaces (APIs) provided by the community media 23, 24.

It is worth mentioning that the main database 206 provided by the server system 20 is larger (or much larger) than the enterprise database 210. In one embodiment, although the enterprise data 210 runs on the same server system 20 as the master database 206, it is preferably two different and independent databases. When the enterprise characteristic model is established according to the enterprise data, the relevance between the main data and various enterprise data is established, so that more consumer lists related to the enterprise attributes can be obtained from the main data, and the enterprise characteristic model is expanded to be an object of enterprise marketing.

FIG. 3 depicts a flowchart illustration of an embodiment of a method of consuming data.

The system establishes master data 302 from various data sources 301 (at least two sources, or more) such as consumption activity in the social media, or data provided from various enterprises. Where primary data 302 is characterized by more dimensionality, much larger than enterprise data.

The enterprise data 303, which encompasses consumer data pertaining to an enterprise attribute, is entered into the system and subjected to preprocessing including data cleansing as described in the previous embodiments to derive valid data, feature acquisition, scoring (304) based on the RFM analysis, and machine learning 305 to learn features in the data, some of which are used for training purposes in order to build models 306. The main data 302 is then input into the model 306, and these steps compensate the deviation and data shortage problem of the enterprise data 303 only for specific enterprise attributes, and obtain the preliminary marketing list 307 with high similarity and similar attributes from the main data 302.

In the consumption data processing method, marketing advertisements (308) are put on all or part of the consumers according to the preliminary marketing list 307, and a marketing result is generated, wherein the marketing result shows whether the consumers in the preliminary marketing list 307 consume or not, and the marketing result becomes the basis for updating the enterprise characteristic model.

For example, according to the marketing result, the consumption behavior of the consumers is set to 1, the consumption behavior of the consumers is set to 0, the main data 302 is subjected to data cleaning and feature acquisition 309, the model 311 is updated by the machine learning 310 method, meanwhile, the volume labels in the data are updated according to the marketing result, the updated marketing list 312 is generated again, the advertisements are put according to the marketing objects again, a new marketing result is obtained similarly, and the enterprise characteristic model is continuously updated according to the marketing result. The results are repeatedly input and retrained to update the enterprise personality model, thereby optimizing the enterprise personality model associated with the enterprise attributes.

Therefore, in the method, when a marketing result is generated, the machine learning algorithm can continuously obtain more features from the main data based on a few features of the original enterprise data according to the marketing result, reestablish a model with more dimensions, and continuously update the optimization model according to the next release.

In the consumption data processing method, a Customer Relationship Management (CRM) analysis may be performed to classify customers in the enterprise data 303 when the enterprise data 303 is subjected to data cleansing, and the customer relationship management analysis is not limited to a specific method, and the RFM analysis method disclosed in the above embodiment is taken as an example, in which the customers in the enterprise data 303 are scored by analyzing the last consumption, consumption frequency and consumption amount so as to obtain the volume label in the enterprise data.

One way to form the master data of the system is to collect consumer consumption preferences and records from various marketing activities, for example, group buying activities initiated by using a community medium, when community users participate in the group buying activities and express interest in or participate in buying group buying objects, a consumer profile (Persona) can be established by the obtained data, the profile is used to describe the information established by the consumers in various consumption behaviors and records, and besides personal basic data, the profile also includes descriptions of consumption preferences and behavior modes.

FIG. 4 depicts a flowchart of one method of establishing master data, which enumerates a method of establishing data through consumption activities of a social media.

In the community medium, the server may recruit members to invite group friends to join a group of group buying activities to commodities provided by suppliers (step S401), propose group buying objects (step S403), and let the group friends express opinions in the community medium with information from and to, including participation in the group buying activities, so that the server may obtain consumer response information through a specific interface (e.g., API) provided by the community medium (step S405), and these response information are analyzed by the server, including performing statistics and semantic analysis (step S407), and may obtain valid data, and establish master data (step S409).

It is noted that the main data created by the consumption data continuously obtained from various sources has relatively wide and complex data, and the sample number is much larger than that of the enterprise data, so that the main data can relatively represent the market parent, and therefore, after the method establishes the association with the main data through intelligently learning the characteristics of the enterprise data, the method can assist the enterprise to effectively expand the marketing objects with originally single attribute to more consumption families.

According to one embodiment, when a member of the system establishes one or more group buying activities in a social media, the user's activities in the social media in the one or more group buying activities may be analyzed by using a chat robot running in a chat room in the social media when analyzing various consumption related information, and analysis results, such as a result of order completion, preferences regarding each group buying activity, and the like, may be generated to establish master data.

According to the embodiment, the servo system can adopt a customer relationship management method to process various consumption related information and is matched with a purchase-sale-inventory system to record commodity data, price, cost, inventory, suppliers and the like in consumption activities. Thus, by collecting member data and consumption data of various retail (such as group purchase) to establish main data, deviation (bias) and deficiency in enterprise data from single attribute can be made up by machine learning, and the enterprise attribute is characterized by high data similarity and fixed attribute or characteristic.

FIG. 5 is a flowchart illustrating a method for processing consumption data by the system.

At first, enterprise data is input into the system (step S501), the enterprise data covers consumer data belonging to a specific enterprise attribute, and compared with the main data of the system, the enterprise data has a large difference and a deviation, and then the enterprise data is subjected to data cleaning and features thereof are obtained (step S503). At this time, a feature profile of each consumer in the enterprise data may be further created, which is a consumer feature profile created according to the records of various consumer activities of the consumer, and may be called a personal character (step S505) for describing the consumption attributes of each consumer.

Further, the above-mentioned RFM analysis is performed from the customer data in the enterprise data after data washing, but not limited to this method, which is intended to be scored and classified (step S507), and these preprocessing procedures generate a training set for subsequently performing machine learning.

It is mentioned here that in the embodiment of the RFM analysis employed for scoring, the last consumption (Recency) index employed refers to the data of the last time the consumer purchased, including the last consumption time and the reference data on which the last consumption is based, which is in principle the consumer with higher score and may also be more responsive to the subsequent marketing information. The consumption Frequency (Frequency) index is the Frequency of consumption of the consumer, and if the Frequency of the consumer is high, the consumer shows that the preference of the goods or the services provided by the enterprise is high, and the consumer has loyalty and is also a high-grade consumer. The consumption amount (money) index adopted refers to the amount consumed in the last period of time, and is a quite intuitive index, the higher the consumption amount is, the higher the willingness of the consumer to consume is, the higher the enterprise profit is, and the important index for grading the consumer is.

Then, a machine learning method is performed on the data subjected to the data cleaning and feature acquisition to acquire the volume label in the enterprise data and establish an enterprise characteristic model (step S509).

And then inputting main data (step S511), establishing the relevance between the enterprise data and the main data, wherein the number of consumers covered by the main data is larger than that of the enterprise data and is not limited by the attributes of the enterprise data, so that a marketing list of characteristics related to the enterprise data can be obtained from the main data (step S513), a target customer group with specific characteristics (aiming at the enterprise) can be obtained, and then marketing advertisements can be put on all or part of the consumers according to the marketing list.

Fig. 6 is a diagram of an embodiment of a method for checking a marketing list.

Once the marketing manifest is generated with the enterprise personality model (step S601), the enterprise personality model is optimized by examining the marketing results, at which point a small number of trial impressions for market verification are made based on the manifest (step S603), resulting in marketing results (step S605), based on which features may be redefined, including again machine learning, to update or build tags in the model based on consumers where the behavioral results match the features (step S607).

Then, an enterprise personality model with main data as the main data is established (step S609), a new marketing list is formed (step S611), and then, the process returns to step S603, if necessary, the enterprise personality model can be repeatedly updated according to the marketing result of each time, and the main data is repeatedly input into the updated enterprise personality model to optimize the enterprise personality model related to the enterprise attributes.

The above-mentioned fig. 4 to fig. 6 describe the flow of the consumption data processing method proposed by the present disclosure, including the establishment of the main data, the main processing flow, and the flow of the subsequent inspection and optimization model, which list the applied scenarios.

An enterprise such as a Thai restaurant, the client group of the Thai restaurant should have the characteristics of enjoying (or not excluding) Thai cuisine, such as people who may enjoy the characteristics of heavy taste, peppery taste, and vegetables. Thus, the database of Thai restaurants includes "customers who like Thai and have high value" who should respond preferentially to future marketing campaigns. According to the method, the characteristics of the "customers who like Thai and have high value" can be extracted by RFM analysis assumption by using the existing data of the Thai restaurant, and the characteristics can be used as the material for machine learning.

By using the method, more main data with wider data to consumers is input into an enterprise characteristic model established according to the enterprise data, more consumers which probably like Thai cooking characteristics, namely other consumers which accord with the characteristics of customers who like Thai and high value can be obtained, at this time, small-amount putting of market verification can be carried out, feature extraction and analysis are carried out according to putting results, more volume labels or features are generated, the group of customers are assumed to be people who like or can accept Thai cooking again, and the model which assumes the customers who like Thai and high value is redefined according to the consumers whose behavior results accord with the features.

Thus, the output results of the established model are more accurate at one time than at one time due to machine learning in the process.

In summary, the embodiment of the consumption data processing method described above is applied to the customer consumption data of the retail industry itself, and such enterprise data is data with almost fixed certain attributes or characteristics according to its industrial characteristics, and therefore has deviation with big data (such as the main data described in the embodiment) formed by multiple sources, and the method can make up for the problem of the deviation and insufficient data through the main data, so that the enterprise can expand potential customer groups other than the original customer group to bring more profits, and the related method flow can also assist the enterprise to develop new customers, new product development and expand the needs of stores.

The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, so that equivalent structural changes made by using the description and drawings of the present invention are included in the claims of the present invention, and it is obvious that the present invention is also covered by the claims.