Information recommendation method, device, computer equipment and computer readable storage medium

Abstract

The present application discloses an information recommendation method, device, computer equipment and computer-readable storage medium, which relates to the field of data processing technology, and screens out common features with high contribution, realizes accurate positioning of users, and improves the accuracy of information recommendation. The method includes: determining the target user to be recommended information, obtaining test data and multiple user features associated with the target user, and dividing the data set composed of the test data and multiple user features into a training set and a verification set; using the training set to train multiple machine learning models, performing contribution analysis based on multiple machine learning models and verification sets, obtaining the contribution of each user feature to the modeling target of each machine learning model, and determining common features among multiple user features; constructing a user positioning model, inputting the common features into the user positioning model, obtaining the user positioning result of the target user, and recommending information to the target user according to the user positioning result.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an information recommendation method, device, computer equipment and computer-readable storage medium. Background Art

[0002] In recent years, with the rapid development of the Internet, modern society has become an information-based and digital society. Data is flooding the world, and information explosion has become the norm. However, in the face of a large amount of data, users' utilization of information has decreased, which has led to the problem of information overload. In this regard, the recommendation system is one of the key technologies to effectively solve the problem of information overload.

[0003] In related technologies, the recommendation system obtains the user's historical behavior data, such as whether the user clicks, purchases, order records, etc., and uses the historical behavior data to build a response model. The response model is used to analyze the user's preferences and habits, thereby positioning the user and making personalized recommendations to the user.

[0004] In the process of implementing the related technology, the applicant realizes that the related technology has at least the following problems:

[0005] When using the constructed response model to analyze user characteristics, all user characteristics of the user need to be input into the response model. However, in fact, when constructing the response model, different user characteristics have different contributions to the modeling target. User characteristics with low contribution can easily affect the output results of the response model, resulting in inaccurate user positioning and low accuracy of information recommendation. Summary of the invention

[0006] In view of this, the present application provides an information recommendation method, apparatus, computer device and computer-readable storage medium, the main purpose of which is to solve the problem that low-contribution user characteristics are likely to affect the results output by the response model, resulting in inaccurate user positioning and low accuracy of information recommendation.

[0007] According to the first aspect of the present application, there is provided an information recommendation method, the method comprising:

[0008] Determine a target user for whom information recommendation is to be performed, obtain test data and a plurality of user features associated with the target user, and divide a data set consisting of the test data and the plurality of user features into a training set and a validation set;

[0009] Using the training set to train multiple machine learning models, performing contribution analysis based on the multiple machine learning models and the validation set to obtain the contribution of each user feature to the modeling target of each machine learning model, and determining common features among the multiple user features, wherein the common features are user features whose contribution to building each machine learning model is greater than a contribution threshold among the multiple user features;

[0010] A user positioning model is constructed, the common features are input into the user positioning model, a user positioning result of the target user is obtained, and information is recommended to the target user according to the user positioning result.

[0011] Optionally, the acquiring the test data and multiple user features associated with the target user, and dividing the data set consisting of the test data and the multiple user features into a training set and a validation set, includes:

[0012] Determine a plurality of initial user features based on the historical behavior data and basic user data of the target user, wherein the historical behavior data is data generated by the historical behavior of the target user and / or data generated by the interaction with the target user within a historical period;

[0013] respectively determining a feature type of each of the plurality of initial user features, and processing the plurality of initial user features according to the feature type to obtain the plurality of user features, wherein the feature type is any one of a numerical feature or a categorical feature;

[0014] Acquire test data associated with the target user, and use the test data and the multiple user features to form the data set, wherein the test data is obtained after testing the target user for a specified target;

[0015] A preset grouping ratio is determined, and the data set is grouped according to the preset grouping ratio to obtain the training set and the validation set.

[0016] Optionally, after determining a plurality of initial user characteristics based on the historical behavior data and basic user data of the target user, the method further includes:

[0017] Identifying an initial user feature whose value is empty among the multiple initial user features as a missing user feature, and identifying a feature type of the missing user feature;

[0018] A type identifier corresponding to the feature type is obtained, and the value of the missing user feature is filled with the type identifier.

[0019] Optionally, the respectively determining a feature type of each of the multiple initial user features, and processing the multiple initial user features according to the feature type to obtain the multiple user features includes:

[0020] Performing feature type identification on the multiple initial user features to determine at least one first initial user feature and at least one second initial user feature, wherein the feature type of the first initial user feature is a numerical feature, and the feature type of the second initial user feature is a categorical feature;

[0021] The at least one first initial user feature is standardized and bucketed, the at least one second initial user feature is label encoded, and the processed at least one first initial user feature and the processed at least one second initial user feature are used as the multiple user features.

[0022] Optionally, the adopting the training set to train multiple machine learning models, performing contribution analysis based on the multiple machine learning models and the validation set, and obtaining the contribution of each user feature to the modeling target of each machine learning model, includes:

[0023] Determine a preset machine learning algorithm, and perform multiple model training operations on the training set according to the preset machine learning algorithm to obtain the multiple machine learning models;

[0024] Obtain a contribution analysis tool, select any one of the multiple machine learning models as a target machine learning model, input the target machine learning model and the validation set into the contribution analysis tool, and obtain the marginal contribution value of each user feature as the contribution of each user feature to the modeling target of the target machine learning model;

[0025] Continue to select any one machine learning model from the remaining machine learning models and input it into the contribution analysis tool together with the verification set until the multiple machine learning models are traversed to obtain the contribution of each user feature to the modeling target of each machine learning model, wherein the remaining machine learning models are machine learning models among the multiple machine learning models except the target machine learning model.

[0026] Optionally, the constructing of the user positioning model includes:

[0027] Acquire test data associated with the target user, and acquire a preset control data group;

[0028] Counting the data difference between the experimental data group and the control data group in the test data on the specified target, wherein the test data is obtained after testing the target user for the specified target;

[0029] A causal model algorithm is determined, and the causal model algorithm is used to perform model training on the data difference to obtain the user positioning model.

[0030] Optionally, the recommending information to the target user according to the user positioning result includes:

[0031] Extracting at least one user location tag from the user location result, respectively querying the recommended content associated with each user location tag in the at least one user location tag to obtain at least one recommended content;

[0032] Counting the number of occurrences of each recommended content in the at least one recommended content, and sorting the at least one recommended content in descending order of the number of occurrences to obtain a content sorting result;

[0033] One or more recommended contents ranked first in the content ranking result are selected as information to be recommended, and the information to be recommended is recommended to the target user.

[0034] According to a second aspect of the present application, there is provided an information recommendation device, the device comprising:

[0035] An acquisition module is used to determine a target user for whom information recommendation is to be performed, acquire test data and a plurality of user features associated with the target user, and divide a data set consisting of the test data and the plurality of user features into a training set and a validation set;

[0036] An analysis module is used to train multiple machine learning models using the training set, perform contribution analysis based on the multiple machine learning models and the validation set to obtain a contribution of each user feature to a modeling target of each machine learning model, and determine a common feature among the multiple user features, wherein the common feature is a user feature among the multiple user features whose contribution to building each machine learning model is greater than a contribution threshold;

[0037] The recommendation module is used to construct a user positioning model, input the common features into the user positioning model, obtain the user positioning result of the target user, and recommend information to the target user according to the user positioning result.

[0038] Optionally, the acquisition module is used to determine multiple initial user features based on the historical behavior data and user basic data of the target user, wherein the historical behavior data is data generated by the historical behavior of the target user and / or data generated by interaction with the target user within a historical period; respectively determine the feature type of each of the multiple initial user features, and process the multiple initial user features according to the feature type to obtain the multiple user features, wherein the feature type is any one of a numerical feature or a categorical feature; obtain test data associated with the target user, and use the test data and the multiple user features to form the data set, wherein the test data is obtained after testing the target user for a specified target; determine a preset grouping ratio, and group the data set according to the preset grouping ratio to obtain the training set and the validation set.

[0039] Optionally, the acquisition module is further used to identify an initial user feature with an empty value as a missing user feature among the multiple initial user features, and to identify a feature type of the missing user feature; obtain a type identifier corresponding to the feature type, and use the type identifier to fill in the value of the missing user feature.

[0040] Optionally, the acquisition module is used to identify the feature types of the multiple initial user features, determine at least one first initial user feature and at least one second initial user feature, the feature type of the first initial user feature is a numerical feature, and the feature type of the second initial user feature is a categorical feature; standardize and bucket the at least one first initial user feature, perform label encoding on the at least one second initial user feature, and use the processed at least one first initial user feature and the processed at least one second initial user feature as the multiple user features.

[0041] Optionally, the analysis module is used to determine a preset machine learning algorithm, and perform multiple model training operations on the training set according to the preset machine learning algorithm to obtain the multiple machine learning models; obtain a contribution analysis tool, select any machine learning model from the multiple machine learning models as the target machine learning model, input the target machine learning model and the validation set into the contribution analysis tool, and obtain the marginal contribution value of each user feature as the contribution of each user feature to the modeling target of the target machine learning model; continue to select any machine learning model from the remaining machine learning models and input it into the contribution analysis tool together with the validation set until the multiple machine learning models are traversed to obtain the contribution of each user feature to the modeling target of each machine learning model, wherein the remaining machine learning models are machine learning models other than the target machine learning model among the multiple machine learning models.

[0042] Optionally, the recommendation module is used to obtain test data associated with the target user and a preset control data group; to count the data differences between the experimental data group and the control data group in the test data on the specified target, and the test data is obtained after testing the target user for the specified target; to determine a causal model algorithm, and to use the causal model algorithm to perform model training on the data differences to obtain the user positioning model.

[0043] Optionally, the recommendation module is used to extract at least one user location tag from the user location result, query the recommended content associated with each user location tag in the at least one user location tag, and obtain at least one recommended content; count the number of occurrences of each recommended content in the at least one recommended content, and sort the at least one recommended content in descending order of the number of occurrences to obtain a content sorting result; select one or more recommended contents ranked first in the content sorting result as the information to be recommended, and recommend the information to be recommended to the target user.

[0044] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0045] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

[0046] By means of the above technical scheme, the present application provides an information recommendation method, device, computer equipment and computer-readable storage medium. The present application determines the target user to be recommended information, obtains test data and multiple user features associated with the target user, and divides the data set composed of the test data and multiple user features into a training set and a verification set, uses the training set to train multiple machine learning models, performs contribution analysis based on multiple machine learning models and the verification set, obtains the contribution of each user feature to the modeling target of each machine learning model, and determines the common features whose contribution to the construction of each machine learning model is greater than the contribution threshold among the multiple user features, constructs a user positioning model, inputs the common features into the user positioning model, obtains the user positioning result of the target user, and recommends information to the target user according to the user positioning result, and through the recognition of contribution, screens out the common features with higher contribution to the modeling target of each machine learning model, and locates the user around these common features to avoid the influence of user features with low contribution on the positioning result, thereby achieving accurate positioning of the user and improving the accuracy of information recommendation.

[0047] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0049] Figure 1 A schematic diagram of a flow chart of an information recommendation method provided in an embodiment of the present application is shown;

[0050] Figure 2A A schematic diagram of another information recommendation method provided in an embodiment of the present application is shown;

[0051] Figure 2B A schematic diagram of the architecture of an information recommendation system provided by an embodiment of the present application is shown;

[0052] Figure 2C A schematic diagram of a flow chart of an information recommendation method provided in an embodiment of the present application is shown;

[0053] Figure 3 A schematic diagram of the structure of an information recommendation device provided in an embodiment of the present application is shown;

[0054] Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0055] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0056] The present application embodiment provides an information recommendation method, such as Figure 1 As shown, the method includes:

[0057] 101. Determine a target user for whom information recommendation is to be performed, obtain test data and multiple user features associated with the target user, and divide a data set consisting of the test data and the multiple user features into a training set and a validation set.

[0058] The embodiment of the present application can be applied to an information recommendation system, which can be installed in APPs (Applications) such as shopping, video, and chat, so as to locate users and recommend content that may be of interest to them according to their preferences. In order to achieve accurate positioning of users, information is subsequently recommended based on the positioning of users. In the embodiment of the present application, after the target user to be recommended is determined, test data and multiple user features associated with the target user are obtained. Among them, the test data can be A / B experimental data, which is obtained after testing the target user for a specified target. In actual applications, specific business goals can be set, and AB tests can be performed for specific business goals to determine which is better, A or B, thereby obtaining test data. Multiple user characteristics can be determined based on the target user's historical behavior data and user basic data; specifically, the historical behavior data may include the target user's outbound call records, historical donation records, etc. The outbound call records mainly include the number of historical connection times, the number of transfers to manual, the connection time, the average waiting time, etc. The historical donation records mainly include the number of donations made by the target user in the past, the number of donation case visits, whether there is a donation record in the past 30 days, etc.; the user basic data is the user's own personal attribute data, which may specifically include the target user's age, occupation, family income, whether social security is paid, etc.

[0059] After obtaining the test data and multiple user features, since the subsequent embodiments of the present application need to analyze the contribution of each user feature to the modeling target of the model, it is necessary to group the test data and multiple user features, one part is used to build the model, and the other part is used to be input into the contribution analysis tool together with the constructed model to perform the contribution analysis of the user features. Among them, the user features used to build the model can be set as the training set, and the user features used to be input into the contribution analysis tool together with the constructed model to perform the contribution analysis of the user features can be set as the verification set. When grouping the test data and multiple user features, they can be grouped according to a preset grouping ratio. For example, if the preset grouping ratio is 7:3, the data set composed of the test data and multiple user features can be divided into 70% and 30%, and the data with a ratio of 7 is used as the training set, and the data with a ratio of 3 is used as the verification set. The present application does not limit the specific value of the preset grouping ratio.

[0060] 102. Use the training set to train multiple machine learning models, perform contribution analysis based on multiple machine learning models and validation sets, obtain the contribution of each user feature to the modeling target of each machine learning model, and determine common features among multiple user features.

[0061] Among them, after completing the division of the training set and the validation set, the training set can be used to train multiple machine learning models. Specifically, the constructed machine learning models can be XGBoost (eXtreme GradientBoosting, extreme gradient boosting), CatBoost (Categorical Features GradientBoosting, category feature gradient boosting), LightGBM (LightGradient Boosting Machine, gradient boosting framework), etc., and this application does not specifically limit this. After constructing multiple machine learning models, in fact, each machine learning model has a modeling goal. For example, when constructing a user positioning model, the modeling goal is to locate the user and determine the user's preferences. In fact, each user feature involved in constructing the user positioning model has a different contribution to the modeling goal. User features related to user preferences, such as historical purchase records and historical browsing records, have a higher contribution to the modeling goal, while user features such as address and user name that are not closely related to user preferences have a lower contribution to the modeling goal. Therefore, the information recommendation system will start to perform contribution analysis based on multiple machine learning models and validation sets to obtain the contribution of each user feature to the construction of each machine learning model, so as to select some user features with higher contribution as the key to positioning the user in the future, so as to achieve accurate positioning of the user. Specifically, when performing contribution analysis, you can use contribution analysis tools such as Shap analysis tools, use the multiple machine learning models and validation sets obtained through training as inputs of the contribution analysis tools, and obtain the contribution of each user feature to the modeling target after contribution analysis processing.

[0062] Next, after obtaining the contribution of each user feature to the modeling target of each machine learning model, since the user features with higher contribution are actually the main factors affecting the information content recommended to the user, it will be more accurate to recommend information to the user based on the set of user features with higher contribution. Therefore, the information recommendation system will determine the common features among multiple user features. The common features are user features whose contribution to the modeling target of each machine learning model is greater than the contribution threshold among multiple user features. Subsequently, information is recommended to the user based on the common features. It should be noted that the common features can be one or more, as long as the user features whose contribution to the modeling target of each machine learning model meets the contribution threshold can be used as common features. Among them, the contribution threshold can be set to 0. In this way, in the embodiment of the present application, the common features whose contribution to the construction of each machine learning model is greater than 0 can be used as at least one common feature.

[0063] 103. Build a user positioning model, input the common features into the user positioning model, obtain the user positioning result of the target user, and recommend information to the target user according to the user positioning result.

[0064] Among them, after the common features are determined, since the common features are user features whose contribution to the modeling target of each machine learning model is greater than the contribution threshold, the accuracy of locating users and recommending information based on these common features is higher. Therefore, in the embodiment of the present application, the information recommendation system will build a user positioning model, input the common features into the user positioning model, and enable the user positioning model to use the common features to locate the user, determine which information the user is interested in, obtain the user positioning result of the target user, and recommend information to the target user according to the user positioning result. Specifically, the user positioning model can be constructed using a causal model algorithm, and the user positioning result can be a result in the form of a label, that is, the user's preferences are expressed with a label. In this way, when recommending information to the target user according to the user positioning result, the information associated with the label can be recommended to the user as the information to be recommended for the user to browse.

[0065] The method provided in the embodiment of the present application determines the target user for information recommendation, obtains test data and multiple user features associated with the target user, and divides the data set composed of the test data and multiple user features into a training set and a validation set, uses the training set to train multiple machine learning models, performs contribution analysis based on the multiple machine learning models and the validation set, obtains the contribution of each user feature to the modeling target of each machine learning model, and determines the common features whose contribution to the construction of each machine learning model is greater than the contribution threshold among the multiple user features, constructs a user positioning model, inputs the common features into the user positioning model, obtains the user positioning result of the target user, and recommends information to the target user according to the user positioning result, screens out the common features with higher contribution to the modeling target of each machine learning model through contribution identification, and locates the user around these common features to avoid the influence of user features with low contribution on the positioning result, thereby achieving accurate positioning of the user and improving the accuracy of information recommendation.

[0066] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, the embodiment of the present application provides another information recommendation method, such as Figure 2A As shown, the method includes:

[0067] 201. Determine a plurality of initial user features based on the historical behavior data and basic user data of the target user.

[0068] The applicant recognizes that in the current personalized recommendation task for users, the usual solution for whether users click or buy is to build a response model, but the defect of this method is that it cannot distinguish between marketing-sensitive people and natural conversion people, and the modeling target and business target are not completely matched; its feature importance is based on the overall data set, which does not meet the requirements of personalized recommendation for users. How to balance the precise positioning of marketing-sensitive people and the realization of explainable modeling is one of the problems that need to be solved to realize personalized recommendation for users. Therefore, the present application proposes an information recommendation method, which determines the contribution of each user feature to the modeling target of each machine learning model through contribution analysis, selects common features with high contribution to the modeling target of each machine learning model, and locates users around these common features, avoids the influence of low-contribution user features on the positioning results, realizes accurate positioning of users, and improves the accuracy of information recommendation; in addition, in the embodiment of the present application, a large amount of A / B experimental data is also used as part of the training machine learning model, and a user positioning model is constructed using a causal model algorithm, which can not only accurately locate marketing-sensitive people, but also analyze individual cases, and can determine the reasons for the difference between the predicted value and the actual value and the contribution of each user feature to the modeling target, so as to meet the requirements of personalized recommendation for users.

[0069] The embodiments of the present application can be applied to an information recommendation system. The information recommendation system can be installed in apps such as shopping, video, and chat, so as to locate users and recommend content that may be of interest to them according to their preferences. The system architecture of the information recommendation system is first introduced below:

[0070] like Figure 2B As shown, the information recommendation system includes a database layer, a data layer, a model layer and a result layer. Among them, the database layer is used to store test data, historical behavior data and basic user data related to users. There is a data link between the data layer and the database layer, which is used to determine the characteristics of users based on the test data, historical behavior data and basic user data provided by the database layer. Since the data provided by the database layer includes the user's own attribute data and the interaction data generated when the user interacts with the APP, the user-related characteristics determined by the data layer will actually include user attribute characteristics and user interaction characteristics. There is a data link between the model layer and the data layer, which is used to analyze the contribution of user characteristics and build a user positioning model. There is a data link between the result layer and the model layer, which is used to locate users using the user positioning model constructed by the model layer and the contribution analysis results, thereby realizing personalized information recommendation for users.

[0071] For the embodiment of the present application, since the database layer stores user-associated test data, historical behavior data, and user basic data, the information recommendation system will determine the target user for information recommendation, determine multiple initial user features based on the target user's historical behavior data and user basic data, and obtain the test data associated with the target user. Among them, historical behavior data is data generated by historical behaviors of target users and / or data generated by interactions with target users in historical time periods. Test data is obtained after testing target users for designated targets. Specifically, test data can be A / B experimental data, which is obtained after testing the target users for designated targets. In actual applications, specific business targets can be set, and AB tests can be conducted for specific business targets to determine which is better, A or B, so as to obtain test data. Historical behavior data can include outbound call records, historical donation records, etc. of target users. Outbound call records mainly include the number of historical connections, the number of transfers to manual, the connection time, the average waiting time, etc. Historical donation records mainly include the number of donations made by target users in the past, the number of donation case visits, whether there are donation records in the past 30 days, etc. User basic data is the user's own personal attribute data, which can specifically include the target user's age, occupation, family income, whether social security is paid, etc. After obtaining the test data, historical behavior data and user basic data associated with the target user, these acquired data can be directly used as multiple initial user features, so that the multiple initial user features can be processed later to determine the multiple user features that are ultimately used for contribution evaluation.

[0072] 202. Determine the feature type of each of the multiple initial user features respectively, and process the multiple initial user features according to the feature type to obtain multiple user features.

[0073] In the embodiment of the present application, each initial user feature corresponds to a feature type, and the feature type is either a numerical feature or a categorical feature. Among them, the numerical feature means that the value of the initial user feature is a number, such as age, number of historical calls, etc.; the categorical feature means that the value of the initial user feature is a certain category, such as gender, which can be divided into male and female categories. When processing these initial user features, the information recommendation system will adopt different processing methods to unify the formats of initial user features of different feature types, so as to facilitate the subsequent unified contribution analysis. The following describes the process of processing multiple initial user features according to feature type:

[0074] Considering that some users' information is not comprehensive enough, there may be missing information, resulting in missing initial user features. For example, some users did not fill in their age, so the value of the user's age feature obtained by the information recommendation system is a null value. For another example, some users did not fill in their gender, so the value of the user's gender feature obtained by the information recommendation system is also a null value. In order to avoid the need for additional special processing of these null values ​​in the subsequent modeling and contribution evaluation process, the information recommendation system will identify the missing initial user features before processing multiple initial user features, and perform missing processing on these missing initial user features. Specifically, the information recommendation system will identify the initial user features with empty values ​​as missing user features among multiple initial user features, identify the feature type of the missing user features, obtain the type identifier corresponding to the feature type, and use the type identifier to fill in the value of the missing user feature. For example, assuming that the type identifier corresponding to the numerical feature is 0 and the type identifier corresponding to the categorical feature is -1, when the missing user feature determined to be age is age, the missing user feature is a numerical feature, and the value of the user's age is set to 0; and when the missing user feature determined to be gender, the missing user feature is a categorical feature, and the value of the user's gender is set to -1.

[0075] After completing the processing of missing values, you can start processing multiple initial user features according to the feature type to obtain multiple user features. Among them, the information recommendation system will identify the feature type of multiple initial user features, determine at least one first initial user feature and at least one second initial user feature, the feature type of the first initial user feature is a numerical feature, and the feature type of the second initial user feature is a categorical feature. Next, standardize and bucket the at least one first initial user feature, perform label encoding on the at least one second initial user feature, and use the processed at least one first initial user feature and the processed at least one second initial user feature as multiple user features. Specifically, the label encoding process for the second initial user feature can be Label-Encoder encoding of the second initial user feature, so as to unify the format of the initial user features of different feature types and obtain multiple user features, which is convenient for subsequent model training and contribution analysis.

[0076] 203. Obtain test data associated with the target user, and use the test data and multiple user features to form a data set.

[0077] The test data can be A / B experimental data, which is obtained after testing the target users for the specified target. In practical applications, specific business goals can be set, and AB tests can be performed for specific business goals to determine which is better, A or B, so as to obtain test data. In the embodiment of the present application, a large amount of A / B experimental data is also used as part of training multiple machine learning models, and a user positioning model is constructed using a causal model algorithm. It can not only accurately locate marketing-sensitive groups, but also analyze individual cases, determine the reasons for the difference between the predicted value and the actual value, and the contribution of each user feature to the modeling target, so as to meet the user's personalized recommendation requirements.

[0078] Therefore, the embodiment of the present application will obtain test data associated with the target user, and use the test data and multiple user features to form a data set, so as to subsequently train multiple machine learning models based on the data set.

[0079] 204. Determine a preset grouping ratio, and group the data set according to the preset grouping ratio to obtain a training set and a validation set.

[0080] In an embodiment of the present application, after obtaining the test data and multiple user features, since the subsequent embodiments of the present application need to analyze the contribution of each user feature to the construction of the model, it is necessary to group the data set composed of the test data and multiple user features, one part is used to build the model, and the other part is used to be input into the contribution analysis tool together with the constructed model to perform the contribution analysis of the user features. Among them, the user features used to build the model can be set as the training set, and the user features used to be input into the contribution analysis tool together with the constructed model to perform the contribution analysis of the user features can be set as the verification set. When grouping the data set, it can be grouped according to the preset grouping ratio. For example, if the preset grouping ratio is 7:3, the data set can be divided into 70% and 30%, and the data with a ratio of 7 is used as the training set, and the data with a ratio of 3 is used as the verification set. The present application does not limit the specific value of the preset grouping ratio.

[0081] 205. Use the training set to train multiple machine learning models, perform contribution analysis based on the multiple machine learning models and the validation set, obtain the contribution of each user feature to the modeling target of each machine learning model, and determine the common features among the multiple user features.

[0082] In the embodiment of the present application, after the division of the training set and the validation set is completed, the training set can be used to train multiple machine learning models. Specifically, the constructed machine learning models can be XGBoost, CatBoost, LightGBM, etc., and the present application does not specifically limit this. After constructing multiple machine learning models, the information recommendation system will start to perform contribution analysis based on multiple machine learning models and validation sets to obtain the contribution of each user feature to the modeling target of each machine learning model. Specifically, when performing contribution analysis, you can use contribution analysis tools such as Shap (SHapley AdditiveexPlanations, machine learning interpretability tool) analysis tools, and use the trained multiple machine learning models and validation sets as inputs to the contribution analysis tool. After the contribution analysis is processed, the contribution of each user feature to the modeling is obtained. The Shap analysis tool uses game theory to explain the output of the machine learning model. When applied to the machine learning model, it can explain how much each feature contributes to the corresponding predicted value. The core concept of the Shap analysis tool is the Shapley value. The Shapley value calculates the marginal contribution of each feature. In layman's terms, the difference between the contribution when this feature is added and the sum of the contributions before this feature is added is the marginal contribution of this feature. It should be noted that there is also a part of the contribution called the base value, which is calculated based on the distribution of the target variable. For regression problems, the base value is the mean of the target variable; for classification problems, the base value is the mean of the encoded target variable. The Shapley value is calculated for each feature, and the final prediction result is the base value "plus" the marginal contribution of all features. The following describes the process of training multiple machine learning models and performing contribution analysis based on multiple machine learning models and validation sets:

[0083] First, the information recommendation system will determine the preset machine learning algorithm, and perform multiple model training operations on the training set according to the preset machine learning algorithm to obtain multiple machine learning models, where the preset machine learning algorithm can be XGBoost, CatBoost, LightGBM and other algorithms.

[0084] Next, the information recommendation system will obtain a contribution analysis tool, select any machine learning model from multiple machine learning models as the target machine learning model, input the target machine learning model and the validation set into the contribution analysis tool, and obtain the marginal contribution value of each user feature as the contribution of each user feature to the modeling target of the target machine learning model. Subsequently, continue to select any machine learning model from the remaining machine learning models and input it into the contribution analysis tool together with the validation set until multiple machine learning models are traversed to obtain the contribution of each user feature to the modeling target of each machine learning model, wherein the remaining machine learning models are machine learning models other than the target machine learning model in the multiple machine learning models. That is to say, the information recommendation system will select a machine learning model each time, input the machine learning model and the validation set into the contribution analysis tool, and then obtain the contribution of each user feature involved in building the machine learning model to the modeling target of the model; next, continue to select other machine learning models to perform the same operation, and also input the machine learning model and the validation set into the contribution analysis tool. In this way, for each user feature, multiple contributions to the modeling targets of multiple machine learning models will be obtained. Subsequently, refer to the multiple contributions corresponding to each user feature, and select user features whose corresponding multiple contributions are greater than the contribution threshold as common features to locate users.

[0085] Next, after obtaining the contribution of each user feature to the modeling target of each machine learning model, since the user features with higher contribution are actually the main factors affecting the information content recommended to the user, it will be more accurate to recommend information to the user with the set of user features with higher contribution as the main guide. Therefore, the information recommendation system will determine the common features among multiple user features. The common features are the user features whose contribution to the construction of each machine learning model is greater than the contribution threshold among multiple user features, and then recommend information to the user based on the common features. Among them, the contribution threshold can be set to 0. In this way, in the embodiment of the present application, the common features whose contribution to the modeling target of each machine learning model is greater than 0 can be used as at least one common feature. For example, assuming that three machine learning models are constructed, namely Model_1, Model_2 and Model_3, the user feature is age, the contribution of age to the modeling target of Model_1 is +5.79, the contribution to the modeling target of Model_2 is +6.08, and the contribution to the modeling target of Model_3 is +4.57, then the user feature of age can be used as a common feature.

[0086] It should be noted that in actual application, if the Shap analysis tool is used to analyze the contribution, the Shap analysis tool will output a Beeswarm graph, which can show the contribution of each user feature to the model target.

[0087] 206. Build a user positioning model, input the common features into the user positioning model, and obtain a user positioning result of the target user.

[0088] Among them, after the common features are determined, since the common features are user features whose contribution to the modeling target of each machine learning model is greater than the contribution threshold, the accuracy of locating users and recommending information based on these common features is higher. Therefore, in the embodiment of the present application, the information recommendation system will build a user positioning model, input the common features into the user positioning model, and enable the user positioning model to use the common features to locate the user, determine which information the user is interested in, and obtain the user positioning result of the target user. The user positioning result can be a result in the form of a label, that is, the user's preferences are expressed with a label. In this way, when recommending information to the target user according to the user positioning result, the information associated with the label can be recommended to the user as the information to be recommended for the user to browse.

[0089] It should be noted that the user positioning model is constructed using a causal model algorithm, and the specific construction process is as follows: First, the information recommendation system obtains the test data associated with the target user, and obtains a preset control data group. Subsequently, the data difference between the experimental data group and the control data group in the test data on the specified target is statistically analyzed, the causal model algorithm is determined, and the causal model algorithm is used to train the model for the data difference to obtain the user positioning model. In other words, the data difference between the experimental data group and the control data group in the test data on the specified target is determined, that is, the uplift increment is determined, and the uplift increment is modeled using the causal model algorithm to obtain a user positioning model that can accurately locate the user.

[0090] 207. Recommend information to target users according to the user positioning results.

[0091] In the embodiment of the present application, after the user positioning result is determined, information can be recommended to the target user according to the user positioning result. Among them, since the user positioning result can be in the form of a tag, the information recommendation system will extract at least one user positioning tag from the user positioning result, and query the recommended content associated with each user positioning tag in the at least one user positioning tag to obtain at least one recommended content. Subsequently, the number of occurrences of each recommended content in at least one recommended content is counted respectively, and at least one recommended content is sorted in order from high to low according to the number of occurrences to obtain a content sorting result. Finally, one or more recommended contents ranked first in the content sorting result are selected as the information to be recommended, and the information to be recommended is recommended to the target user. Assuming that the user positioning tags are tags 1, 2, and 3, the recommended content obtained by tag 1 is A, B, and C, the recommended content obtained by tag 2 is A, C, and D, and the recommended content obtained by tag 3 is A, E, and F, then the sorting result obtained by sorting the recommended content according to the number of occurrences is A, C, B, E, and F, and A or A and C can be recommended to the user as the information to be recommended for the user to browse.

[0092] In summary, the logical process of the technical solution of this application is summarized as follows: Figure 2C As shown, multiple initial user features of the target user are obtained, and the multiple initial user features may specifically include user attribute features and user interaction features. Next, the feature type of each initial user feature is identified, and it is determined which are numerical features and which are categorical features. The numerical features are standardized and bucketed, and the categorical features are label encoded to obtain multiple user features. Subsequently, the data set consisting of multiple user features of the test data associated with the target user is divided into a training set and a validation set, and multiple machine learning models are trained using the training set. Contribution analysis is performed based on multiple machine learning models and validation sets to obtain the contribution of each user feature to the modeling target of each machine learning model, and common features are determined among multiple user features. These common features are input into the user positioning model constructed using the causal model algorithm to achieve user positioning and information recommendation.

[0093] In this way, this application uses causal inference and contribution analysis to replace the traditional response model. On the one hand, it solves the problem that the response model cannot distinguish between naturally converted people and marketing sensitive people, cannot analyze individual cases, and cannot identify the contribution of each feature to the modeling target, and does not meet the requirements of personalized recommendations; on the other hand, compared with the response model, it can bring more significant benefits while reducing operating costs.

[0094] The method provided in the embodiment of the present application determines the target user for information recommendation, obtains test data and multiple user features associated with the target user, and divides the data set composed of the test data and multiple user features into a training set and a validation set, uses the training set to train multiple machine learning models, performs contribution analysis based on the multiple machine learning models and the validation set, obtains the contribution of each user feature to the modeling target of each machine learning model, and determines the common features whose contribution to the construction of each machine learning model is greater than the contribution threshold among the multiple user features, constructs a user positioning model, inputs the common features into the user positioning model, obtains the user positioning result of the target user, and recommends information to the target user according to the user positioning result, screens out the common features with higher contribution to the modeling target of each machine learning model through contribution identification, and locates the user around these common features to avoid the influence of user features with low contribution on the positioning result, thereby achieving accurate positioning of the user and improving the accuracy of information recommendation.

[0095] Further, as Figure 1 In a specific implementation of the method, the present application provides an information recommendation device, such as Figure 3 As shown, the device includes: an acquisition module 301, an analysis module 302 and a recommendation module 303.

[0096] The acquisition module 301 is used to determine the target user for information recommendation, obtain the test data and multiple user features associated with the target user, and divide the data set consisting of the test data and the multiple user features into a training set and a validation set;

[0097] The analysis module 302 is used to train multiple machine learning models using the training set, perform contribution analysis based on the multiple machine learning models and the validation set to obtain the contribution of each user feature to the modeling target of each machine learning model, and determine common features among the multiple user features, wherein the common features are user features whose contribution to building each machine learning model is greater than a contribution threshold among the multiple user features;

[0098] The recommendation module 303 is used to construct a user positioning model, input the common features into the user positioning model, obtain the user positioning result of the target user, and recommend information to the target user according to the user positioning result.

[0099] In a specific application scenario, the acquisition module 301 is used to determine multiple initial user features based on the historical behavior data and user basic data of the target user, wherein the historical behavior data is data generated by the historical behavior of the target user and / or data generated by the interaction with the target user in a historical period; respectively determine the feature type of each of the multiple initial user features, and process the multiple initial user features according to the feature type to obtain the multiple user features, wherein the feature type is any one of a numerical feature or a categorical feature; obtain test data associated with the target user, and use the test data and the multiple user features to form the data set, wherein the test data is obtained after testing the target user for a specified target; determine a preset grouping ratio, and group the data set according to the preset grouping ratio to obtain the training set and the verification set.

[0100] In a specific application scenario, the acquisition module 301 is also used to identify an initial user feature with an empty value as a missing user feature among the multiple initial user features, and to identify the feature type of the missing user feature; obtain a type identifier corresponding to the feature type, and use the type identifier to fill in the value of the missing user feature.

[0101] In a specific application scenario, the acquisition module 301 is used to identify the feature types of the multiple initial user features, determine at least one first initial user feature and at least one second initial user feature, the feature type of the first initial user feature is a numerical feature, and the feature type of the second initial user feature is a categorical feature; standardize and bucket the at least one first initial user feature, perform label encoding on the at least one second initial user feature, and use the processed at least one first initial user feature and the processed at least one second initial user feature as the multiple user features.

[0102] In a specific application scenario, the analysis module 302 is used to determine a preset machine learning algorithm, perform multiple model training operations on the training set according to the preset machine learning algorithm, and obtain the multiple machine learning models; obtain a contribution analysis tool, select any machine learning model from the multiple machine learning models as the target machine learning model, input the target machine learning model and the validation set into the contribution analysis tool, and obtain the marginal contribution value of each user feature as the contribution of each user feature to the modeling target of the target machine learning model; continue to select any machine learning model from the remaining machine learning models and input it into the contribution analysis tool together with the validation set until the multiple machine learning models are traversed to obtain the contribution of each user feature to the modeling target of each machine learning model, wherein the remaining machine learning models are machine learning models other than the target machine learning model among the multiple machine learning models.

[0103] In a specific application scenario, the recommendation module 303 is used to obtain the test data associated with the target user and to obtain a preset control data group; to count the data differences between the experimental data group and the control data group in the test data on the specified target, wherein the test data is obtained after testing the target user for the specified target; to determine a causal model algorithm, and to use the causal model algorithm to perform model training on the data differences to obtain the user positioning model.

[0104] In a specific application scenario, the recommendation module 303 is used to extract at least one user location tag from the user location result, query the recommended content associated with each user location tag in the at least one user location tag, and obtain at least one recommended content; count the number of occurrences of each recommended content in the at least one recommended content, and sort the at least one recommended content in descending order of the number of occurrences to obtain a content sorting result; select one or more recommended contents ranked first in the content sorting result as the information to be recommended, and recommend the information to be recommended to the target user.

[0105] The device provided in the embodiment of the present application determines the target user to whom information recommendation is to be made, obtains test data and multiple user features associated with the target user, and divides the data set composed of the test data and multiple user features into a training set and a validation set, uses the training set to train multiple machine learning models, performs contribution analysis based on the multiple machine learning models and the validation set, obtains the contribution of each user feature to the modeling target of each machine learning model, and determines the common features whose contribution to the construction of each machine learning model is greater than the contribution threshold among the multiple user features, constructs a user positioning model, inputs the common features into the user positioning model, obtains the user positioning result of the target user, and recommends information to the target user according to the user positioning result, screens out the common features with higher contribution to the modeling target of each machine learning model through contribution identification, and locates the user around these common features to avoid the influence of user features with low contribution on the positioning result, thereby achieving accurate positioning of the user and improving the accuracy of information recommendation.

[0106] It should be noted that for other corresponding descriptions of the functional units involved in the information recommendation device provided in the embodiment of the present application, reference can be made to Figure 1 and FIG. 2A to FIG. 2C The corresponding description in will not be repeated here.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0108] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

[0110] In an exemplary embodiment, see Figure 4, and also provides a computer device, which includes a bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device, wherein each functional unit can communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the information recommendation method in the above embodiment.

[0111] A computer-readable storage medium stores a computer program, which implements the steps of the information recommendation method when executed by a processor.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by hardware, or by means of software plus necessary general hardware platforms. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0113] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present application.

[0114] Those skilled in the art will appreciate that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0115] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0116] The above disclosure only discloses several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by technicians in this field should fall within the scope of protection of the present application.

Claims

1. An information recommendation method, characterized in that: include: Determine a target user for whom information recommendation is to be performed, obtain test data and a plurality of user features associated with the target user, and divide a data set consisting of the test data and the plurality of user features into a training set and a validation set; Using the training set to train multiple machine learning models, performing contribution analysis based on the multiple machine learning models and the validation set to obtain the contribution of each user feature to the modeling target of each machine learning model, and determining common features among the multiple user features, wherein the common features are user features whose contribution to building each machine learning model is greater than a contribution threshold among the multiple user features; Obtaining a preset control data group, and counting the data difference between the experimental data group and the control data group in the test data on the specified target, wherein the test data is obtained after testing the target user for the specified target; Determine a causal model algorithm, use the causal model algorithm to perform model training on the data difference, build a user positioning model, input the common features into the user positioning model, and obtain a user positioning result of the target user; At least one user location tag is extracted from the user location result, and the recommended content associated with each user location tag in the at least one user location tag is queried respectively to obtain at least one recommended content; the number of occurrences of each recommended content in the at least one recommended content is counted respectively, and the at least one recommended content is sorted in descending order according to the number of occurrences to obtain a content sorting result; one or more recommended contents ranked first in the content sorting result are selected as the information to be recommended, and the information to be recommended is recommended to the target user.

2. The method according to claim 1, characterized in that The obtaining of the test data and a plurality of user features associated with the target user, and dividing a data set consisting of the test data and the plurality of user features into a training set and a validation set, includes: Determine a plurality of initial user features based on the historical behavior data and basic user data of the target user, wherein the historical behavior data is data generated by the historical behavior of the target user and / or data generated by the interaction with the target user within a historical period; respectively determining a feature type of each of the plurality of initial user features, and processing the plurality of initial user features according to the feature type to obtain the plurality of user features, wherein the feature type is any one of a numerical feature or a categorical feature; Acquire test data associated with the target user, and use the test data and the multiple user features to form the data set, wherein the test data is obtained after testing the target user for a specified target; A preset grouping ratio is determined, and the data set is grouped according to the preset grouping ratio to obtain the training set and the validation set.

3. The method according to claim 2, characterized in that After determining a plurality of initial user characteristics based on the historical behavior data and basic user data of the target user, the method further includes: Identifying an initial user feature whose value is empty among the multiple initial user features as a missing user feature, and identifying a feature type of the missing user feature; A type identifier corresponding to the feature type is obtained, and the value of the missing user feature is filled with the type identifier.

4. The method according to claim 2, characterized in that: The step of respectively determining the feature type of each of the multiple initial user features, and processing the multiple initial user features according to the feature type to obtain the multiple user features includes: Performing feature type identification on the multiple initial user features to determine at least one first initial user feature and at least one second initial user feature, wherein the feature type of the first initial user feature is a numerical feature, and the feature type of the second initial user feature is a categorical feature; The at least one first initial user feature is standardized and bucketed, the at least one second initial user feature is label encoded, and the processed at least one first initial user feature and the processed at least one second initial user feature are used as the multiple user features.

5. The method according to claim 1, characterized in that The using the training set to train multiple machine learning models, performing contribution analysis based on the multiple machine learning models and the validation set, and obtaining the contribution of each user feature to the modeling target of each machine learning model, includes: Determine a preset machine learning algorithm, and perform multiple model training operations on the training set according to the preset machine learning algorithm to obtain the multiple machine learning models; Obtain a contribution analysis tool, select any one of the multiple machine learning models as a target machine learning model, input the target machine learning model and the validation set into the contribution analysis tool, and obtain the marginal contribution value of each user feature as the contribution of each user feature to the modeling target of the target machine learning model; Continue to select any one machine learning model from the remaining machine learning models and input it into the contribution analysis tool together with the verification set until the multiple machine learning models are traversed to obtain the contribution of each user feature to the modeling target of each machine learning model, wherein the remaining machine learning models are machine learning models among the multiple machine learning models except the target machine learning model.

6. An information recommendation device, characterized in that: include: An acquisition module is used to determine a target user for whom information recommendation is to be performed, acquire test data and a plurality of user features associated with the target user, and divide a data set consisting of the test data and the plurality of user features into a training set and a validation set; An analysis module is used to train multiple machine learning models using the training set, perform contribution analysis based on the multiple machine learning models and the validation set to obtain a contribution of each user feature to a modeling target of each machine learning model, and determine a common feature among the multiple user features, wherein the common feature is a user feature among the multiple user features whose contribution to building each machine learning model is greater than a contribution threshold; A recommendation module is used to obtain a preset control data group, count the data differences between the experimental data group and the control data group in the test data on the specified target, wherein the test data is obtained after testing the target user for the specified target; determine a causal model algorithm, use the causal model algorithm to perform model training on the data differences, build a user positioning model, input the common features into the user positioning model, and obtain a user positioning result for the target user; extract at least one user positioning tag from the user positioning result, and query the recommended content associated with each user positioning tag in the at least one user positioning tag to obtain at least one recommended content; count the number of occurrences of each recommended content in the at least one recommended content, and sort the at least one recommended content in descending order of the number of occurrences to obtain a content sorting result; select one or more recommended contents ranked first in the content sorting result as information to be recommended, and recommend the information to be recommended to the target user.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images