Training method, usage method, device, equipment and medium of feature reconstruction model

Abstract

The present application discloses a training method, usage method, device, equipment and medium of a feature reconstruction model, belonging to the field of artificial intelligence technology. The method includes: obtaining the first general knowledge feature of the first account, and obtaining the second general knowledge feature and the auxiliary feature of the second account; performing encoding processing on the first general knowledge feature to obtain the first reconstructed general knowledge feature, performing encoding processing on the second general knowledge feature to obtain the second reconstructed general knowledge feature, and performing encoding processing on the auxiliary feature to obtain the reconstructed auxiliary feature; predicting the account types to which the first reconstructed general knowledge feature and the second reconstructed general knowledge feature belong to obtain a predicted classification result; training the feature reconstruction model according to the comparison error and the adversarial error. By training the feature reconstruction model, the present application narrows the difference between the accounts of the first type and the second type of account types; avoids the problem of difficult clustering caused by the information difference between the accounts of different types.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and particularly to a training method, a usage method, a device, a device and a medium for a feature reconstruction model. Background Art

[0002] With the development of Internet technology, the demand for clustering different users and providing personalized services according to different clustering results is increasing.

[0003] In the related art, usually, by extracting the behavior information of users in the application, such as the browsing information of users on multimedia information and the usage information of users on the application, the users are clustered according to the behavior information to obtain the clustering results of the users.

[0004] However, this method relies heavily on the behavior information of users in the application. For users who use the application for the first time or users with little behavior information in the application, it is often difficult to classify. How to improve the clustering results is an urgent problem to be solved. Summary of the Invention

[0005] The present application provides a training method, a usage method, a device, a device and a medium for a feature reconstruction model, and the technical solutions are as follows:

[0006] According to one aspect of the present application, a training method for a feature reconstruction model is provided, and the method includes:

[0007] Obtain the first general feature of the first account, and obtain the second general feature and the auxiliary feature of the second account. The first general feature includes the feature representation of the first general information of the first account, the second general feature includes the feature representation of the second general information of the second account, and the auxiliary feature includes the feature representation of the auxiliary information of the second account. The account type of the first account is the first type, the account type of the second account is the second type, and there is no corresponding auxiliary information for the accounts of the first type.

[0008] Perform encoding processing on the first general feature to obtain a first reconstructed general feature, perform encoding processing on the second general feature to obtain a second reconstructed general feature, and perform encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature.

[0009] Predict the account types to which the first reconstructed general feature and the second reconstructed general feature belong to obtain a predicted classification result.

[0010] Train the feature reconstruction model according to the contrast error and the adversarial error to obtain a trained feature reconstruction model. The contrast error includes the error between the second reconstructed general feature and the reconstructed auxiliary feature. The adversarial error includes the error between the predicted classification result and the actual classification result. The actual classification result is the account type actually corresponding to the first general feature and the second general feature.

[0011] According to another aspect of the present application, there is provided a method for using a feature reconstruction model. The feature reconstruction model is trained according to the training method of the above feature reconstruction model. The method includes:

[0012] Obtain the general features of at least one third account. The general features include the feature representation of the general information of the third account.

[0013] Perform encoding processing on the general features based on the feature reconstruction model to obtain reconstructed general features.

[0014] Perform clustering processing on the reconstructed general features to obtain the clustering result of the third account.

[0015] According to another aspect of the present application, there is provided an information recommendation method. The method includes:

[0016] Obtain the application features of multiple target application accounts. The target application accounts include at least one fourth account with an account type of inactive type and at least one fifth account with an account type of active type. The application features include the feature representation of the application information of the target application accounts. The application information includes the installation and uninstallation information of at least one application by the target application accounts.

[0017] Perform encoding processing on the application features based on a feature reconstruction model. The feature reconstruction model is trained according to the training method of the above feature reconstruction model.

[0018] Perform clustering processing on the reconstructed application features to obtain the clustering result of the target application accounts.

[0019] Based on the information of the fifth account, determine the information recommendation information of the fourth account that belongs to the same clustering result as the fifth account. The information includes the information browsing information of the fifth account in the target application.

[0020] According to another aspect of the present application, there is provided a training device for a feature reconstruction model. The device includes:

[0021] An acquisition module, configured to acquire a first general feature of a first account, and acquire a second general feature and an auxiliary feature of a second account, where the first general feature includes a feature representation of first general information of the first account, the second general feature includes a feature representation of second general information of the second account, and the auxiliary feature includes a feature representation of auxiliary information of the second account; the account type of the first account is a first type, the account type of the second account is a second type, and there is no corresponding auxiliary information for the accounts of the first type.

[0022] An encoding module, configured to perform encoding processing on the first general feature to obtain a first reconstructed general feature, perform encoding processing on the second general feature to obtain a second reconstructed general feature, and perform encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature;

[0023] A prediction module, configured to predict the account types to which the first reconstructed general feature and the second reconstructed general feature belong, and obtain a predicted classification result;

[0024] A training module, configured to train the feature reconstruction model according to a contrast error and an adversarial error, and obtain a trained feature reconstruction model, where the contrast error includes an error between the second reconstructed general feature and the reconstructed auxiliary feature, the adversarial error includes an error between the predicted classification result and an actual classification result, and the actual classification result is the actual corresponding account type of the first general feature and the second general feature.

[0025] In an alternative design of the present application, the feature reconstruction model includes a contrast learning encoder;

[0026] The encoding module is further configured to:

[0027] Call the contrast learning encoder to perform encoding processing on the first general feature to obtain the first reconstructed general feature, call the contrast learning encoder to perform encoding processing on the second general feature to obtain the second reconstructed general feature, and call the contrast learning encoder to perform encoding processing on the auxiliary feature to obtain the reconstructed auxiliary feature.

[0028] In an alternative design of the present application, the feature reconstruction model further includes a general encoder and an auxiliary encoder;

[0029] The acquisition module is further configured to:

[0030] Acquire the first general information of the first account, and acquire the second general information and the auxiliary information of the second account;

[0031] Call the general encoder to encode the first general information to obtain the first general feature, call the general encoder to encode the second general information to obtain the second general feature, and call the auxiliary encoder to encode the auxiliary information to obtain the auxiliary feature.

[0032] In an alternative design of the present application, the first general information and the second general information are multi-hot encoded vectors, and the auxiliary information is a continuous value vector.

[0033] In an alternative design of the present application, the feature reconstruction model further includes a general decoder and an auxiliary decoder;

[0034] The apparatus further includes:

[0035] A decoding module, configured to call the general decoder to decode the first general feature to obtain a first reconstructed general information, call the general decoder to decode the second general feature to obtain a second reconstructed general information, and call the auxiliary decoder to decode the auxiliary feature to obtain a reconstructed auxiliary information;

[0036] The training module is further configured to:

[0037] Train the feature reconstruction model according to the contrast error, the adversarial error, and the reconstruction error to obtain the trained feature reconstruction model;

[0038] Wherein, the reconstruction error includes at least one of the error between the first reconstructed general information and the first general information, the error between the second reconstructed general information and the second general information, and the error between the reconstructed auxiliary information and the auxiliary information.

[0039] In an alternative design of the present application, at least one of the contrast error, the adversarial error, and the reconstruction error includes a cross-entropy loss.

[0040] In an alternative design of the present application, the feature reconstruction model includes a gradient reversal layer and an account classifier;

[0041] The prediction module is further configured to:

[0042] Call the gradient reversal layer to perform a reversal process on the first reconstructed general feature and the second reconstructed general feature to obtain a processed first reconstructed general feature and a processed second reconstructed general feature;

[0043] Call the account classifier to predict the account types of the processed first reconstructed general feature and the processed second reconstructed general feature to obtain the predicted classification result.

[0044] In an alternative design of the present application, the first general information is first application information, the first application information includes installation and uninstallation information of at least one application for the first account, the second general information is second application information, the second application information includes installation and uninstallation information of at least one application for the second account, the auxiliary information is information, and the information includes information browsing information of the second account in the target application;

[0045] The obtaining module is further configured to:

[0046] Obtain a first application feature of the first account, and obtain a second application feature and an information feature of the second account, where the first application feature includes a feature representation of the first application information of the first account, the second application feature includes a feature representation of the second application information of the second account, and the information feature includes a feature representation of the information of the second account;

[0047] The encoding module is further configured to:

[0048] Perform encoding processing on the first application feature to obtain a first reconstructed application feature, perform encoding processing on the second application feature to obtain a second reconstructed application feature, and perform encoding processing on the information feature to obtain a reconstructed information feature;

[0049] The prediction module is further configured to:

[0050] Predict the account types to which the first reconstructed application feature and the second reconstructed application feature belong to obtain the predicted classification result.

[0051] According to another aspect of the present application, there is provided a device for using a feature reconstruction model, the device includes:

[0052] An obtaining module, configured to obtain general features of at least one third account, where the general features include feature representations of general information of the third account;

[0053] An encoding module, configured to perform encoding processing on the general features based on the feature reconstruction model to obtain reconstructed general features;

[0054] A clustering module, configured to perform clustering processing on the reconstructed general features to obtain a clustering result of the third account.

[0055] In an alternative design of the present application, the device further includes:

[0056] A determination module, configured to obtain auxiliary information of the third account within a target time period, and determine account features of a target clustering cluster according to the auxiliary information, where the target clustering cluster includes the clustering cluster indicated by the clustering result of the third account.

[0057] In an alternative design of the present application, the determination module is further configured to:

[0058] Call a sorting model to perform sorting processing on the account features to obtain recommendation information of the accounts to be recommended, where the accounts to be recommended belong to the target clustering cluster.

[0059] In an alternative design of the present application, the device further includes:

[0060] A determination module, configured to obtain high-frequency auxiliary information of the third account exceeding a target threshold within a target time period, and determine recommendation information of the accounts to be recommended according to the high-frequency auxiliary information, where the third account and the accounts to be recommended belong to the target clustering cluster, and the target clustering cluster includes the clustering cluster indicated by the clustering result of the third account.

[0061] According to another aspect of the present application, there is provided an information recommendation device, where the device includes:

[0062] An acquisition module, configured to acquire application features of multiple target application accounts, where the target application accounts include at least one fourth account with an account type of inactive type and at least one fifth account with an account type of active type, and the application features include a feature representation of application information of the target application accounts, and the application information includes installation and uninstallation information of the target application accounts for at least one application;

[0063] An encoding module, configured to perform encoding processing on the application features based on a feature reconstruction model to obtain reconstructed application features;

[0064] A clustering module, configured to perform clustering processing on the reconstructed application features to obtain a clustering result of the target application accounts;

[0065] A determination module, configured to determine information recommendation information of the fourth accounts belonging to the same clustering result as the fifth account based on the information of the fifth account, where the information includes information browsing information of the fifth account in the target application.

[0066] According to another aspect of the present application, there is provided a computer device, where the computer device includes a processor and a memory, and at least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method and / or usage method and / or information recommendation method of the feature reconstruction model as described in the above aspects.

[0067] According to another aspect of the present application, there is provided a computer-readable storage medium storing at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the method for reconstructing a feature model and / or the usage method and / or the information recommendation method as described in the above aspect.

[0068] According to another aspect of the present application, there is provided a computer program product including computer instructions stored in a computer-readable storage medium, and a processor reads and executes the computer instructions from the computer-readable storage medium to implement the method for reconstructing a feature model and / or the usage method and / or the information recommendation method as described in the above aspect.

[0069] The beneficial effects brought by the technical solution provided by the present application at least include:

[0070] By reconstructing features through encoding processing, the difference between general knowledge information and auxiliary information is reduced, and by predicting to obtain a predicted classification result, the difference between accounts of the first type and the second type of account types is reduced; training the feature reconstruction model according to the comparison error and the adversarial error realizes the alignment of accounts of the first type and the second type of account types, eliminates the influence of auxiliary information on the second general knowledge information of the second account during the clustering process, ensures the effect of clustering accounts, and avoids the problem of difficult clustering caused by information differences between accounts of different types. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0072] Figure 1 is a schematic diagram of a computer system provided by an exemplary embodiment of the present application;

[0073] Figure 2 is a schematic diagram of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0074] Figure 3 is a schematic diagram of a method for using a feature reconstruction model provided by an exemplary embodiment of the present application;

[0075] Figure 4It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0076] Figure 5 It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0077] Figure 6 It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0078] Figure 7 It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0079] Figure 8 It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0080] Figure 9 It is a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application;

[0081] Figure 10 It is a flowchart of a method for using a feature reconstruction model provided by an exemplary embodiment of the present application;

[0082] Figure 11 It is a flowchart of an information recommendation method provided by an exemplary embodiment of the present application;

[0083] Figure 12 It is a flowchart of an information recommendation method provided by an exemplary embodiment of the present application;

[0084] Figure 13 It is an interface diagram of recommended information provided by an exemplary embodiment of the present application;

[0085] Figure 14 It is a flowchart of an information recommendation method provided by an exemplary embodiment of the present application;

[0086] Figure 15 It is a schematic diagram of an account clustering model provided by an exemplary embodiment of the present application;

[0087] Figure 16 It is a structural block diagram of a training device for a feature reconstruction model provided by an exemplary embodiment of the present application;

[0088] Figure 17 It is a structural block diagram of a using device for a feature reconstruction model provided by an exemplary embodiment of the present application;

[0089] Figure 18 It is a structural block diagram of an information recommendation device provided by an exemplary embodiment of the present application;

[0090] Figure 19 It is a structural block diagram of a server provided by an exemplary embodiment of the present application.

[0091] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Detailed implementation manners

[0092] To make the objectives, technical solutions, and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0093] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0094] The terms used in this disclosure are for the purpose of describing particular embodiments only and are not intended to limit the disclosure. The singular forms "a", "the", and "said" used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0095] 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 for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions. For example, the first general information of the first account, the second general information of the second account, and the auxiliary information involved in the present application are all obtained under the full authorization of the user.

[0096] It should be understood that although the terms first, second, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this disclosure, the first parameter may also be referred to as the second parameter, and similarly, the second parameter may also be referred to as the first parameter. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0097] Figure 1 The figure shows a schematic diagram of a computer system provided by an embodiment of the present application. The computer system can be implemented as a system architecture for a training method and / or a usage method of a feature reconstruction model. The computer system may include: a terminal 100 and a server 200. The terminal 100 can be an electronic device such as a mobile phone, a tablet computer, an in-vehicle terminal (car computer), a wearable device, a PC (Personal Computer), an unmanned reservation terminal, etc. A client for running a target application can be installed in the terminal 100. The target application can be a training and / or usage application for a feature reconstruction model, or other applications provided with the training and / or usage function of the feature reconstruction model. The present application does not make any limitation in this regard. In addition, the present application does not limit the form of the target application, including but not limited to an App (Application) installed in the terminal 100, a mini-program, etc., and can also be in the form of a web page. The server 200 can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The server 200 can be the background server of the above target application, and is used to provide background services for the client of the target application.

[0098] For the training method and / or the usage method of the feature reconstruction model provided by the embodiments of the present application, the execution entity of each step can be a computer device, and the computer device refers to an electronic device with data calculation, processing, and storage capabilities. Taking Figure 1 the solution implementation environment shown as an example, the training method and / or the usage method of the feature reconstruction model can be executed by the terminal 100 (for example, the client of the target application installed and running in the terminal 100 executes the training method and / or the usage method of the feature reconstruction model), or the training method and / or the usage method of the feature reconstruction model can be executed by the server 200, or the terminal 100 and the server 200 can interact and cooperate to execute. The present application does not make any limitation in this regard.

[0099] In addition, the technical solution of the present application can be combined with blockchain technology. For example, for the training method and / or the usage method of the feature reconstruction model disclosed in the present application, some of the data involved can be saved on the blockchain. The terminal 100 and the server 200 can communicate with each other through a network, such as a wired or wireless network.

[0100] In the related art, it is necessary to perform clustering processing on different accounts and provide personalized services according to different clustering results. Since the account types are divided into active types and inactive types, the behavior information of accounts of different account types in the application is completely different, and there are domain differences between active-type and inactive-type accounts.

[0101] For example, taking the information indicating the multimedia information browsing behavior of an account within an application as an example, an account of the active type has corresponding information, while an account of the inactive type does not have corresponding information. Further, the application information of the account is used to indicate the installation and uninstallation information of at least one application; the information corresponding to the active type of account will affect the application information; for example, an account that frequently browses multimedia information installs more niche application programs.

[0102] In the process of clustering accounts based on the behavior information of the accounts in the application, due to the domain differences between the active type and inactive type of accounts, the clustering result of the accounts is not good; in one implementation, the clustering process in the related art clusters the active type and inactive type of accounts into different clusters respectively.

[0103] In this application, in view of the domain differences between the active type and inactive type of accounts, by narrowing the difference between the information and application information of the active type of accounts; and by narrowing the difference between the application information of the active type of accounts and the application information of the inactive type of accounts, the influence of the domain differences between the active type and inactive type of accounts on account clustering is eliminated, ensuring the effect of clustering the accounts.

[0104] Next, the feature reconstruction model in this application will be introduced:

[0105] Figure 2 FIG. shows a schematic diagram of a training method for a feature reconstruction model provided by an embodiment of this application.

[0106] The feature reconstruction model 330 includes: an encoding network 330a and a classification network 330b.

[0107] Among them, the encoding network 330a includes: a general encoder 332a, an auxiliary encoder 332b, and a contrastive learning encoder 332c; exemplarily, in this embodiment, the encoding network 330a further includes: a general decoder 338a and an auxiliary decoder 338b.

[0108] The classification network 330b includes: a gradient reversal layer 334 and an account classifier 336.

[0109] Exemplarily, obtain the first application information 312 of the first account 310, and obtain the second application information 322 and information 324 of the second account 320.

[0110] Exemplarily, the first application information 312 includes the installation and uninstallation information of at least one application by the first account 310, the second application information 322 includes the installation and uninstallation information of at least one application by the second account 320, and the information 324 includes the information browsing information of the second account 320 in the target application.

[0111] Among them, the account type of the first account 310 is an inactive type, and the account type of the second account 320 is an active type; there is no corresponding information for an account of the inactive type.

[0112] Invoke the general encoder 332a to perform encoding processing on the first application information 312 to obtain the first application feature 312a; invoke the general encoder 332a to perform encoding processing on the second application information 322 to obtain the second application feature 322a.

[0113] Invoke the auxiliary encoder 332b to perform encoding processing on the information 324 to obtain the information feature 324a.

[0114] Invoke the contrastive learning encoder 332c to perform encoding processing on the first application feature 312a to obtain the first reconstructed application feature 312b; invoke the contrastive learning encoder 332c to perform encoding processing on the second application feature 322a to obtain the second reconstructed application feature 322b; invoke the contrastive learning encoder 332c to perform encoding processing on the information feature 324a to obtain the reconstructed information feature 324b.

[0115] Exemplarily, this embodiment further includes:

[0116] Invoke the general decoder 338a to perform encoding processing on the first application feature 312a to obtain the first reconstructed application information 312d; invoke the general decoder 338a to perform encoding processing on the second application feature 322a to obtain the second reconstructed application information 322d.

[0117] Invoke the auxiliary decoder 338b to perform encoding processing on the information feature 324a to obtain the reconstructed information 324d.

[0118] Invoke the classification network 330b to predict the account types to which the first reconstructed application feature 312b and the second reconstructed application feature 322b belong.

[0119] Specifically, the gradient reversal layer 334 is called to reverse process the first reconstructed application feature 312b and the second reconstructed application feature 322b, obtaining the processed first reconstructed application feature and the processed second reconstructed application feature; the account classifier 336 is called to predict the account types of the processed first reconstructed application feature and the processed second reconstructed application feature, obtaining the first predicted classification result 312c of the first reconstructed application feature 312b and the second predicted classification result 322c of the second reconstructed application feature 322b.

[0120] Based on the comparison error between the second reconstructed application feature 322b and the reconstructed information feature 324b, the adversarial error between the first predicted classification result 312c, the second predicted classification result 322c and the actual classification result, and the reconstruction error between the first application information 312 and the first reconstructed application information 312d, the second application information 322 and the second reconstructed application information 322d, and the information 324 and the reconstructed information 324d; the feature reconstruction model 330 is trained to obtain the trained feature reconstruction model.

[0121] Among them, the actual classification result is the account type actually corresponding to the first application information 312 and the second application information 322.

[0122] Figure 3 The figure shows a schematic diagram of the usage method of the feature reconstruction model provided by an embodiment of the present application.

[0123] The trained feature reconstruction model 420 includes an encoding network 420a; specifically, the encoding network 420a includes a trained general encoder 422a and a trained contrastive learning encoder 422b.

[0124] Obtain the application information 412 of at least one third account, and the application information 412 includes the installation and uninstallation information of the at least one application by the third account.

[0125] Call the trained feature reconstruction model 420 to perform encoding processing on the application information 412, obtaining the reconstructed application feature 412b.

[0126] Specifically, call the trained general encoder 422a to perform encoding processing on the application information 412, obtaining the application feature 412a; call the trained contrastive learning encoder 422b to perform encoding processing on the application feature 412a, obtaining the reconstructed application feature 412b.

[0127] By calling the clustering network 432 to perform clustering processing on the reconstructed application feature 412b, obtaining the clustering result 412c of at least one third account.

[0128] Exemplarily, the clustering result 412c of at least one third account is used to indicate at least one clustering cluster, and at least one account with the same account feature is included in one clustering cluster. For example: at least one account with an anime preference corresponds to the first clustering cluster, and at least one account with a news preference corresponds to the second clustering cluster; it should be noted that the above description of the clustering cluster is only an exemplary introduction, and the accounts corresponding to the clustering cluster have the same account features. The account features can directly indicate preference features, or can be just an abstract account feature without indicating the preference content of the account.

[0129] In order to improve the feature reconstruction effect of the feature reconstruction model, it is necessary to train the feature reconstruction model. Next, the training method of the feature reconstruction model will be introduced through the following embodiments.

[0130] Figure 4 The flowchart of the training method of the feature reconstruction model provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. This method includes:

[0131] Step 510: Obtain the first general feature of the first account, and obtain the second general feature and auxiliary feature of the second account;

[0132] The first general feature includes the feature representation of the first general information of the first account. Exemplarily, the first general information is the attribute information of the first account, or the information having a binding relationship with the first account; for example, the information determined or obtained when registering the first account. In one example, each first account respectively corresponds to corresponding first general information, and the first account also corresponds to the first general information without using or accessing the first account.

[0133] The second general feature includes the feature representation of the second general information of the second account, and the auxiliary feature includes the feature representation of the auxiliary information of the second account. Exemplarily, the second general information is the attribute information of the second account, or the information having a binding relationship with the second account. Exemplarily, the auxiliary information is the information generated during the application process of the second account; for example, the information determined or obtained when using or accessing the second account.

[0134] Exemplarily, the manifestation forms of the feature representation include but are not limited to at least one of a feature vector, a feature matrix, a feature value, or bit information.

[0135] The account type of the first account is the first type, and the account type of the second account is the second type; wherein, the information richness of the accounts in the first type is lower than that of the accounts in the second type, and there is no corresponding auxiliary information for the accounts in the first type. Exemplarily, the first type is an inactive type, and the second type is an active type.

[0136] Exemplarily, the account in the first type usually has no corresponding auxiliary information due to fewer account behaviors than the second account; however, it does not exclude the situation where the account in the first type has no corresponding auxiliary information due to the inability to obtain the corresponding auxiliary information.

[0137] Step 520: Perform encoding processing on the first general feature to obtain a first reconstructed general feature, perform encoding processing on the second general feature to obtain a second reconstructed general feature, and perform encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature;

[0138] Exemplarily, the encoding processing can be achieved by prediction through a feature reconstruction model or by statistical calculation through a feature reconstruction model. This embodiment does not make any restrictions on this.

[0139] Exemplarily, the presentation forms of the first reconstructed general feature and the first general feature can be the same or different, and both include but are not limited to at least one of a feature vector, a feature matrix, a feature value, or bit information. Similarly, the presentation forms of the second general feature, the second reconstructed general feature, the auxiliary feature, and the reconstructed auxiliary feature can be the same or different.

[0140] It should be noted that the encoding processing in this step can be executed simultaneously or sequentially in any order; that is, this embodiment does not make any restrictive regulations on the timing relationship of encoding the first general feature, the second general feature, and the auxiliary feature.

[0141] Step 530: Predict the account types to which the first reconstructed general feature and the second reconstructed general feature belong to obtain a predicted classification result;

[0142] Exemplarily, the predicted classification result indicates the account types to which the predicted first reconstructed general feature and second reconstructed general feature belong. Exemplarily, the predicted classification result includes probability information of the first type and / or the second type.

[0143] Step 540: Train the feature reconstruction model according to the contrast error and the adversarial error to obtain a trained feature reconstruction model;

[0144] Exemplarily, the contrast error includes the error between the second reconstructed general feature and the reconstructed auxiliary feature.

[0145] The adversarial error includes the error between the predicted classification result and the actual classification result, and the actual classification result is the actual account type corresponding to the first general feature and the second general feature.

[0146] Exemplarily, the adversarial error is used to guide the trained feature reconstruction model to be unable to predict the actual classification result. For example: the trained feature reconstruction model is unable to accurately predict the account types to which the first reconstructed general feature and the second reconstructed general feature belong.

[0147] Exemplarily, by training the feature reconstruction model with the contrast error and the adversarial error, the difference in describing the second general information and the auxiliary information of the second account through the feature representation is reduced, and the difference in describing the first general information of the first account and the second general information of the second account through the feature representation is reduced, that is, the accounts of the first type and the second type are aligned.

[0148] In summary, the method provided in this embodiment reconstructs the feature through encoding processing, reduces the difference between the general information and the auxiliary information, obtains the predicted classification result through prediction, and reduces the difference between the accounts of the first type and the second type; trains the feature reconstruction model according to the contrast error and the adversarial error, realizes the alignment of the accounts of the first type and the second type, eliminates the influence of the auxiliary information on the second general information of the second account during the clustering process, ensures the effect of clustering the accounts, and avoids the difficult clustering problem caused by the information difference between the accounts of different types.

[0149] Figure 5 The flowchart of the training method of the feature reconstruction model provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. That is, in Figure 4 In the illustrated embodiment, step 520 can be implemented as step 522:

[0150] Step 522: Call the contrastive learning encoder to perform encoding processing on the first general feature to obtain the first reconstructed general feature, call the contrastive learning encoder to perform encoding processing on the second general feature to obtain the second reconstructed general feature, and call the contrastive learning encoder to perform encoding processing on the auxiliary feature to obtain the reconstructed auxiliary feature;

[0151] Exemplarily, the feature reconstruction model includes a contrastive learning encoder. Among them, the contrastive learning encoder includes at least one of the following networks: Convolutional Neural Networks (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM).

[0152] Exemplarily, the encoding processing processes of the contrastive learning encoder are independent of each other; specifically:

[0153] h(v) = f con (z (v) ; θ c ));

[0154] Wherein, f con represents the contrastive learning encoder, and θ c represents the parameters of the contrastive learning encoder; z (v) represents a feature representation, and h (v) represents the reconstructed feature corresponding to the feature representation.

[0155] For example: when v = 1, z (1) represents the first general feature or the second general feature, and h (1) represents the corresponding first reconstructed general feature or the second reconstructed general feature.

[0156] When v = 2, z (2) represents the auxiliary feature, and h (2) represents the corresponding reconstructed auxiliary feature.

[0157] In an alternative implementation, at least one of the first reconstructed general feature, the second reconstructed general feature, and the reconstructed auxiliary feature is a continuous value vector. In one implementation, the continuous value vector is also referred to as a dense vector (DenseVector); it can be understood that the dimensions of the first reconstructed general feature, the second reconstructed general feature, and the reconstructed auxiliary feature are usually the same, but there may also be cases where the dimensions are different.

[0158] In summary, the method provided in this embodiment expands the construction method of the feature reconstruction model by calling the contrastive learning encoder for encoding processing, and reduces the difference between the general information and the auxiliary information through the feature representation encoded by the contrastive learning encoder; it ensures the effect of clustering the accounts and avoids the problem of difficult clustering caused by the information differences between different types of accounts.

[0159] Next, the detailed content of the contrastive error is introduced:

[0160] Exemplarily, the contrastive error is used to describe the error between the second reconstructed general feature and the reconstructed auxiliary feature; it can be understood that the contrastive error includes but is not limited to at least one of the following: Cross-Entropy Loss, Zero-One Loss, Dice Loss.

[0161] In an example, the contrastive error includes:

[0162]

[0163]

[0164] Among them, L con represents the comparison error, U a represents the number of second accounts, and τ represents the temperature hyperparameter of contrastive learning. represents and the cosine similarity between represents the second reconstructed general feature of the i-th second account, represents the reconstructed auxiliary feature of the j-th second account.

[0165] It should be noted that, in an example, the comparison error is used to describe the error between the second reconstructed general feature and the reconstructed auxiliary feature; the feature reconstruction model is trained with the comparison error, and the encoding parameters when obtaining the second reconstructed general feature and the reconstructed auxiliary feature through encoding processing are adjusted. Optionally, the second reconstructed general feature is obtained by encoding the second general feature through calling a contrastive learning encoder, and the reconstructed auxiliary feature is obtained by encoding the auxiliary feature through calling a contrastive learning encoder.

[0166] Exemplarily, the comparison error reduces the difference between the second reconstructed general feature and the reconstructed auxiliary feature, and trains the feature reconstruction model; the comparison error is used to eliminate the domain difference between the auxiliary information of the second account and the general information of the second account, and the comparison error is used to eliminate the influence of the auxiliary information of the second account on the second general information of the second account; the comparison error aligns the second general information and the auxiliary information.

[0167] Furthermore, the feature reconstruction model is trained with the comparison error, and the encoding parameters of the contrastive learning encoder are adjusted; and the contrastive learning encoder encodes the first general feature to obtain the first reconstructed general feature; the contrastive learning encoder encodes the second general feature to obtain the second reconstructed general feature. The comparison error eliminates the domain difference between the auxiliary information of the second account and the general information of the second account, laying a foundation for reducing the domain information carried in the first reconstructed general feature and the second reconstructed general feature and eliminating the domain difference caused by different account types between the first account and the second account when predicting the account type.

[0168] Further, for the first account, the account type of the first account is an inactive type, and there is no corresponding auxiliary information for the first account. While the account type of the second account is an active type, and there is corresponding auxiliary information for the second account; that is, there is a difference in information types between the first account and the second account; further, the auxiliary information of the second account affects the general knowledge information of the second account. Even if the auxiliary information of the second account is directly deleted, there is still a difference caused by different information types between the first general knowledge information of the first account and the second general knowledge information of the second account.

[0169] In one example, the second general knowledge information and the auxiliary information of the second account are aligned by training the feature reconstruction model through a contrast error; and both the first reconstructed general feature and the second reconstructed general feature are obtained through encoding and processing by a contrast learning encoder. The contrast error is used to align the second general knowledge information and the auxiliary information, and the contrast error is used to eliminate the difference in information types between the first account and the second account. The contrast error is used to eliminate the interference caused by different information types between the first account and the second account during the feature reconstruction process. The contrast error is used to eliminate the influence caused by different information types between the first account and the second account during the process of clustering accounts based on the reconstructed features. The contrast error is used to ensure that the second general knowledge information of the second account is not affected by the auxiliary information, and to ensure that the first general knowledge information of the first account and the second general knowledge information of the second account belong to the same domain.

[0170] Figure 6 The flowchart of the training method of the feature reconstruction model provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. That is, in Figure 5 In the shown embodiment, step 510 can be implemented as step 512 and step 514:

[0171] Step 512: Obtain the first general knowledge information of the first account, and obtain the second general knowledge information and the auxiliary information of the second account;

[0172] The account type of the first account is the first type, the account type of the second account is the second type, and there is no corresponding auxiliary information for the accounts of the first type.

[0173] Exemplarily, the first general knowledge information corresponds to the first account, and the second general knowledge information corresponds to the second general knowledge information; the general knowledge information includes the first general knowledge information and the second general knowledge information, and the general knowledge information is the common information that has a corresponding relationship with both the first account and the second account. The auxiliary information corresponds to the second account, and the general knowledge information is the individual information that has a corresponding relationship only with the second account.

[0174] In an alternative implementation, the first general knowledge information and the second general knowledge information are multi-hot encoding vectors (Multi-Hot Vector); specifically:

[0175]

[0176]

[0177] Among them, represents the first general knowledge information, represents the m-th dimensional content in the first general knowledge information, and the vector dimension of the first general knowledge information is M-dimensional; similarly, represents the second general knowledge information, represents the m-th dimensional content in the second general knowledge information, and the vector dimension of the second general knowledge information is M-dimensional.

[0178] It can be understood that in an example, the dimensions of the first general knowledge information and the second general knowledge information are the same, but the case where the dimensions are different is not excluded.

[0179] In an optional implementation manner, the auxiliary information is a continuous value vector; in an implementation manner, the continuous value vector is also called a dense vector; specifically:

[0180]

[0181] Among them, represents the auxiliary information, R represents a real number, and R N represents that the auxiliary information is a vector composed of N-dimensional real numbers.

[0182] It can be understood that the dimension N of the auxiliary information and the dimension M of the first general knowledge information and the second general knowledge information may be the same or different, and this embodiment does not make any restrictions on this.

[0183] In an optional implementation manner, step 512 can be implemented as: obtaining the first general knowledge information of the first account, and obtaining the second general knowledge information and the auxiliary information of the second account with the separate consent of the authorized object;

[0184] Those skilled in the art can understand that in one implementation of the present application, the acquisition of the first general knowledge information, the second general knowledge information, and the auxiliary information has obtained the separate consent of the authorized entity and complies with the relevant laws, regulations, and standards of the relevant countries and regions. Exemplarily, the authorized entity is clearly prompted by at least one of the following methods: privacy agreement, pop-up prompt, etc., and is obtained after the full authorization of the authorized entity; in one implementation, the authorized entity is prompted during each process of the above information acquisition, and the separate consent of the authorized entity is obtained. In one implementation, when requesting the authorized entity to obtain the above information, the purpose of the information is shown to the user. The acquisition of the above information will not be used beyond the scope of the user's consent, and the above information will not be uploaded, stored, or made public without the full authorization of the authorized entity.

[0185] Step 514: Invoke the general knowledge encoder to perform encoding processing on the first general knowledge information to obtain the first general knowledge feature, invoke the general knowledge encoder to perform encoding processing on the second general knowledge information to obtain the second general knowledge feature, and invoke the auxiliary encoder to perform encoding processing on the auxiliary information to obtain the auxiliary feature;

[0186] Exemplarily, the feature reconstruction model further includes a general knowledge encoder and an auxiliary encoder. Similarly, the general knowledge encoder and the auxiliary encoder include but are not limited to at least one of the following networks: convolutional neural network, recurrent neural network, long short-term memory network.

[0187] Exemplarily, the encoding processing processes of the general knowledge encoder are independent of each other; specifically:

[0188]

[0189] where x (v) represents the information of the first account or the second account, E (v) represents the encoder, represents the parameters corresponding to the encoder, and z (v) represents the feature of the first account or the second account.

[0190] For example:

[0191]

[0192]

[0193]

[0194] where E (1) represents the general knowledge encoder, represents the parameters of the general knowledge encoder; represents the first general knowledge information, represents the first general knowledge feature; represents the second general knowledge information, Represents the second general feature.

[0195] E (2) Represents an auxiliary encoder, Represents the parameters of the auxiliary encoder; Represents auxiliary information, Represents an auxiliary feature.

[0196] It should be noted that the encoding process in this step can be executed simultaneously or in any order; that is, this embodiment does not make any restrictive regulations on the timing relationship of encoding the first general information, the second general information, and the auxiliary information.

[0197] In summary, the method provided in this embodiment expands the construction method of the feature reconstruction model by calling the general encoder and the auxiliary encoder, ensures the effect of clustering the accounts, and avoids the difficulty of clustering caused by the information differences between different types of accounts.

[0198] Figure 7 Shows a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application. This method can be executed by a computer device. That is, in Figure 6 In the illustrated embodiment, step 540 can be implemented as step 542, and further includes step 535:

[0199] Step 535: Call the general decoder to perform decoding processing on the first general feature to obtain the first reconstructed general information, call the general decoder to perform decoding processing on the second general feature to obtain the second reconstructed general information, and call the auxiliary decoder to perform decoding processing on the auxiliary feature to obtain the reconstructed auxiliary information;

[0200] Exemplarily, the feature reconstruction model further includes a general decoder and an auxiliary decoder.

[0201] Exemplarily, the decoding processes of the general decoder are independent of each other; specifically:

[0202]

[0203] Among them, x' (v) Represents the reconstructed information of the first account or the second account, D (v) Represents the decoder, Represents the parameters corresponding to the decoder, z (v) Represents the feature of the first account or the second account.

[0204] For example:

[0205]

[0206]

[0207]

[0208] Among them, D (1) represents a general decoder, represents the parameters of the general decoder; represents the first reconstructed general information, represents the first general feature; represents the second reconstructed general information, represents the second general feature.

[0209] D (2) represents an auxiliary decoder, represents the parameters of the auxiliary decoder; represents the reconstructed auxiliary information, represents the auxiliary feature.

[0210] It should be noted that the decoding process in this step can be executed simultaneously or sequentially in any order; that is, this embodiment does not make any restrictive regulations on the timing relationship of decoding the first general feature, the second general feature, and the auxiliary feature.

[0211] Step 542: Train the feature reconstruction model according to the contrast error, the adversarial error, and the reconstruction error to obtain the trained feature reconstruction model;

[0212] Exemplarily, the reconstruction error includes at least one of the error between the first reconstructed general information and the first general information, the error between the second reconstructed general information and the second general information, and the error between the reconstructed auxiliary information and the auxiliary information.

[0213] In an alternative implementation, at least one of the contrast error, the adversarial error, and the reconstruction error includes cross-entropy loss.

[0214] In an alternative implementation, train the feature reconstruction model according to the sum of the contrast error, the adversarial error, and the reconstruction error to obtain the trained feature reconstruction model.

[0215] For example:

[0216] L = L rec + L con + L adv ;

[0217] Among them, L represents the sum of the contrast error, the adversarial error, and the reconstruction error, L con represents the contrast error, L adv represents the adversarial error, L rec represents the reconstruction error.

[0218] It should be noted that in this embodiment, step 535 can be executed before, after, or simultaneously with any step in the first branch, and the first branch includes step 522 and step 530; this embodiment does not make any restrictive regulations on the timing relationship between step 535 and the first branch.

[0219] In summary, the method provided in this embodiment expands the construction method of the feature reconstruction model by calling the general decoder and the auxiliary decoder for decoding processing, and trains the feature reconstruction model through the reconstruction error, ensuring the effect of clustering the accounts and avoiding the difficult clustering problem caused by the information differences between different types of accounts.

[0220] Next, the detailed content of the reconstruction error will be introduced:

[0221] In one example, the reconstruction error includes the error between the first reconstructed general information and the first general information, the error between the second reconstructed general information and the second general information, and the error between the reconstructed auxiliary information and the auxiliary information.

[0222] It can be understood that the reconstruction error includes but is not limited to at least one of the following: cross-entropy loss function, 0-1 loss function, Dice loss function.

[0223] In one example, the reconstruction error includes:

[0224]

[0225]

[0226]

[0227] Among them, L rec represents the reconstruction error, U represents the sum of the number of the first account and the second account, M represents the dimension of the first reconstructed general information and the second reconstructed general information, represents the m-th dimension content of the general information of the i-th account, represents the m-th dimension content of the reconstructed general information of the i-th account, U a represents the number of the second account, represents the auxiliary information of the i-th second account, represents the reconstructed auxiliary information of the i-th second account, ||||2 represents the 2-norm operation, and σ represents the Sigmoid function.

[0228] Figure 8 shows a flowchart of a training method for a feature reconstruction model provided by an exemplary embodiment of the present application. This method can be executed by a computer device. That is, in Figure 4In the illustrated embodiment, step 530 may be implemented as step 532 and step 534:

[0229] Step 532: Invoke the gradient reversal layer to perform inversion processing on the first reconstructed general feature and the second reconstructed general feature, obtaining the processed first reconstructed general feature and the processed second reconstructed general feature;

[0230] Exemplarily, the feature reconstruction model includes a gradient reversal layer and an account classifier.

[0231] Exemplarily, align the first reconstructed general feature of the first account and the second reconstructed general feature of the second account through the gradient reversal layer (Gradient Reversal Layer, GRL) and the account classifier; by training the feature reconstruction model, it is achieved that the second reconstructed general feature does not carry information specific to the domain.

[0232] Exemplarily, the gradient reversal layer includes:

[0233] R λ (χ) = χ;

[0234]

[0235] where I represents the identity matrix, λ represents the meta-parameter, and χ represents the first reconstructed general feature or the second reconstructed general feature; the gradient reversal layer has no parameters associated with it, and the meta-parameter will not be updated through backpropagation. During the forward propagation process, the gradient reversal layer plays the role of an identity transformation.

[0236] Step 534: Invoke the account classifier to predict the account types of the processed first reconstructed general feature and the processed second reconstructed general feature, obtaining the predicted classification result;

[0237] Exemplarily, the account classifier includes at least one of the following: Support Vector Machine (SVM), Naive Bayes classifier, decision tree, Logistic regression.

[0238] Exemplarily, the predicted classification result includes:

[0239] p active (α|h (1) ) = softmax(f adv (α|h (1) ; φ));

[0240] where h (1) represents the reconstructed general feature, and p active (·|h (1)) represents the predicted classification result, that is, the probability that the reconstructed general feature belongs to the target account category. α represents the target account category. Exemplarily, the target account category is the first type corresponding to the first account or the second type corresponding to the second account.

[0241] Among them, softmax represents the normalization method function, f adv represents the account classifier, and φ represents the parameters of the account classifier.

[0242] Furthermore, the adversarial error is introduced:

[0243] Exemplarily, the adversarial error includes the error between the predicted classification result and the actual classification result; it can be understood that the comparison error includes at least one of the following: cross-entropy loss function, 0-1 loss function, Dice loss function.

[0244] In one example, the adversarial error includes:

[0245]

[0246] Among them, L adv represents the adversarial error, U represents the sum of the numbers of the first account and the second account, y i represents the actual classification result of the i-th account, represents the reconstructed general feature of the i-th account; represents the probability that the account type belongs to the actual classification result of the i-th account under the condition of the reconstructed general feature of the i-th account.

[0247] It should be noted that, in one example, the comparison error is used to describe the error between the predicted classification result and the actual classification result; the feature reconstruction model is trained through the adversarial error, and the encoding parameters when obtaining the first reconstructed general feature and the second reconstructed general feature by encoding processing are adjusted; and the prediction parameters when predicting the account type are adjusted. Optionally, the first general feature is encoded by calling the contrastive learning encoder to obtain the first reconstructed general feature, the second general feature is encoded by calling the contrastive learning encoder to obtain the second reconstructed general feature, and the account classifier is called to predict the account types of the processed first reconstructed general feature and the processed second reconstructed general feature to obtain the predicted classification result.

[0248] Exemplarily, the adversarial error reduces the domain information carried in the first reconstructed general feature and the second reconstructed general feature, and trains the feature reconstruction model; the adversarial error is used to eliminate the domain differences caused by different account types between the first account and the second account; the adversarial error is used to align the first general information of the first account and the second general information of the second account.

[0249] In summary, the method provided in this embodiment performs prediction processing by calling the gradient reversal layer and the account classifier, expands the construction method of the feature reconstruction model, and through the gradient reversal layer, it guides the trained feature reconstruction model to be unable to predict the actual classification result, ensuring the effect of clustering accounts and avoiding the difficult clustering problem caused by the information differences between accounts of different types.

[0250] Figure 9 FIG. shows a flowchart of a method for training a feature reconstruction model provided by an exemplary embodiment of the present application. This method can be executed by a computer device. That is, in Figure 4 In the illustrated embodiment, step 510 can be implemented as step 510a, step 520 can be implemented as step 520a, and step 530 can be implemented as step 530a:

[0251] Step 510a: Obtain the first application feature of the first account, and obtain the second application feature and information feature of the second account;

[0252] In this embodiment, the first general information is the first application information, the first application information includes the installation and uninstallation information of at least one application by the first account, the second general information is the second application information, the second application information includes the installation and uninstallation information of at least one application by the second account, and the auxiliary information is the information information, the information information includes the information browsing information of the second account in the target application.

[0253] Exemplarily, at least one application usually includes the target application, but it does not exclude the case where the target application is an application other than at least one application.

[0254] Exemplarily, among the installation and uninstallation information of at least one application by the first account, at least one application includes at least one of the following: an application logged in with the first account, an application logged in with a bound account having a binding relationship with the first account, and an application logged in with a computer device having a binding relationship with the first account.

[0255] Exemplarily, the first application information and the second application information are multi-hot encoding vectors (Multi-Hot Vector); the vector dimensions of the first application information and the second application information are M dimensions. Among them, the content of the m-th dimension corresponds to the installation and uninstallation information of the m-th application.

[0256] Furthermore, when the m-th application is installed in the first account, the content of the m-th dimension of the first application information is equal to 1, otherwise the content of the m-th dimension of the first application information is equal to 0.

[0257] Exemplarily, the information is a continuous value vector, and the information behavior of the second account is represented by a continuous value vector; optionally, the information is a concatenation of average pre-trained embeddings of clicked and unclicked information.

[0258] Exemplarily, the information includes at least one of the following information: articles, card information, videos, audios, text information, blog posts, multimedia information. Exemplarily, the information affects the second application information of the second account. For example, an account with a high frequency of multimedia information browsing behavior installs more niche application programs.

[0259] Among them, the first application feature includes a feature representation of the first application information of the first account, the second application feature includes a feature representation of the second application information of the second account, and the information feature includes a feature representation of the information of the second account.

[0260] Step 520a: Perform encoding processing on the first application feature to obtain a first reconstructed application feature, perform encoding processing on the second application feature to obtain a second reconstructed application feature, and perform encoding processing on the information feature to obtain a reconstructed information feature;

[0261] Exemplarily, the encoding processing in this step can be executed simultaneously or sequentially in any order; that is, this embodiment does not make any restrictive regulations on the timing relationship of encoding processing the first application feature, the second application feature, and the information feature.

[0262] Step 530a: Predict the account types to which the first reconstructed application feature and the second reconstructed application feature belong to obtain a predicted classification result;

[0263] Exemplarily, the predicted classification result indicates the account types to which the predicted first application general feature and the second reconstructed application feature belong.

[0264] In summary, the method provided in this embodiment ensures the effect of clustering accounts and improves the information recommendation effect for accounts in the application by determining the specific implementation manners of the first general information, the second general information, and the auxiliary information.

[0265] Those skilled in the art can understand that in another implementation manner, the first general information, the second general information, and the auxiliary information at least have the following implementation manners:

[0266] In one implementation, the first general information includes first sales information, the second general information includes second sales information, and the auxiliary information includes after-sales service information; the first sales information is at least one purchase information of the first account, the second sales information is at least one purchase information of the second account, and the after-sales service information is the after-sales service information corresponding to at least one purchase information of the second account. Exemplarily, the after-sales service information affects the second sales information of the second account. For example, the after-sales service information corresponds to a first timestamp, and the after-sales service information affects the information after the first timestamp in the second sales information.

[0267] Specifically, obtain the first sales feature of the first account, and obtain the second sales feature and service feature of the second account; perform encoding processing on the first sales feature to obtain a first reconstructed sales feature, perform encoding processing on the second sales feature to obtain a second reconstructed sales feature, and perform encoding processing on the service feature to obtain a reconstructed service feature; predict the account type to which the first reconstructed sales feature and the second reconstructed sales feature belong to obtain a predicted classification result; train the feature reconstruction model according to the comparison error and the adversarial error to obtain a trained feature reconstruction model.

[0268] In one implementation, the first general information includes first visitor information, the second general information includes second visitor information, and the auxiliary information includes registered member information; the first visitor information is at least one visitor information accessing the first account, the second visitor information is at least one visitor information accessing the second account, and the registered member information is at least one visitor information accessing the second account among the registered members. Exemplarily, the registered member information affects the second visitor information of the second account. For example, for the visitors of the second account, after there are registered members, the registered members of the second account affect the second visitor information of the second account.

[0269] Specifically, obtain the first visitor feature of the first account, and obtain the second visitor feature and member feature of the second account; perform encoding processing on the first visitor feature to obtain a first reconstructed visitor feature, perform encoding processing on the second visitor feature to obtain a second reconstructed visitor feature, and perform encoding processing on the member feature to obtain a reconstructed member feature; predict the account type to which the first reconstructed visitor feature and the second reconstructed visitor feature belong to obtain a predicted classification result; train the feature reconstruction model according to the comparison error and the adversarial error to obtain a trained feature reconstruction model.

[0270] Figure 10 The flowchart of the usage method of the feature reconstruction model provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. This method includes:

[0271] Step 610: Obtain the general features of at least one third account;

[0272] Exemplarily, the general feature includes a feature representation of the general information of the third account.

[0273] Exemplarily, the representation forms of the feature representation include, but are not limited to, at least one of a feature vector, a feature matrix, an eigenvalue, or bit information.

[0274] The account type of the third account can be the first type or the second type; exemplarily, the first type is an inactive type and the second type is an active type.

[0275] Step 620: Perform encoding processing on the general feature based on the feature reconstruction model to obtain a reconstructed general feature;

[0276] Exemplarily, the feature reconstruction model is obtained by the training method of any of the above feature reconstruction models.

[0277] In an alternative implementation, the feature reconstruction model includes a contrastive learning encoder. Call the contrastive learning encoder to perform encoding processing on the general feature to obtain a reconstructed general feature.

[0278] Further optionally, the feature reconstruction model further includes a general encoder, and the general feature of the third account is obtained by calling the general encoder to encode the general information of the third account.

[0279] Step 630: Perform clustering processing on the reconstructed general feature to obtain a clustering result of the third account;

[0280] Exemplarily, the clustering processing is implemented by calling a clustering model; wherein the clustering model is constructed using at least one of K-Means clustering, hierarchical clustering, and density clustering.

[0281] Exemplarily, the clustering result is used to indicate at least one clustering cluster, and at least one third account with the same account features is included in one clustering cluster. Further, the accounts in one clustering cluster have the same preference, and such preference can include specific preference content; for example: the accounts in the first clustering cluster all have an anime preference, and the accounts in the second clustering cluster all have a news preference. It can be understood that the accounts in the clustering cluster having the same preference can also be abstract preference content, which is only used to indicate that the accounts in the clustering cluster belong to the same clustering cluster.

[0282] Optionally, in one implementation, the method further includes:

[0283] Obtain the auxiliary information of the third account within the target time period, and determine the account features of the target clustering cluster according to the auxiliary information;

[0284] The target clustering cluster includes the clustering cluster indicated by the clustering result of the third account. Exemplarily, the account features of the target clustering cluster include the portrait of the account group of the target clustering cluster.

[0285] Further optionally, the method further includes: calling a sorting model to perform sorting processing on the account features to obtain the recommendation information of the account to be recommended.

[0286] The account to be recommended belongs to the target clustering cluster. Adding the account features of the target clustering cluster, that is, adding the portrait of the account group of the target clustering cluster, and obtaining the sorting information of the recommendation information of the account to be recommended through the sorting model.

[0287] Optionally, in another implementation, the method further includes:

[0288] Obtaining the high-frequency auxiliary information of the third account that exceeds the target threshold within the target time period, and determining the recommendation information of the account to be recommended according to the high-frequency auxiliary information;

[0289] The third account and the account to be recommended belong to the target clustering cluster, and the target clustering cluster includes the clustering cluster indicated by the clustering result of the third account. Exemplarily, the high-frequency auxiliary information is the information whose quantity or generation frequency exceeds the target threshold during the application process of the third account within the target time period.

[0290] In summary, the method provided in this embodiment reconstructs the general knowledge features through the feature reconstruction model, and performs clustering processing on the reconstructed general knowledge features, ensuring the effect of clustering the accounts, and avoiding the difficult clustering problem caused by the information differences between different types of accounts.

[0291] Figure 11 The flowchart of the information recommendation method provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. The method includes:

[0292] Step 660: Obtain the application features of multiple target application accounts;

[0293] The application features include the feature representation of the application information of the target application account, and the application information includes the installation and uninstallation information of at least one application by the target application account.

[0294] Exemplarily, the account types of the target application accounts include active types and inactive types. In the case where the account type is an active type, the account also corresponds to information, and the information is the information browsing information of the account in the target application. In the case where the account type is an inactive type, there is no corresponding information for this account.

[0295] Exemplarily, the target application accounts include at least one fourth account with an account type of inactive type and at least one fifth account with an account type of active type.

[0296] Step 670: Encoding the application features based on the feature reconstruction model to obtain reconstructed application features;

[0297] Exemplarily, the feature reconstruction model is obtained by the training method of any of the above feature reconstruction models.

[0298] It should be noted that in this step, the application features of multiple target application accounts are encoded. During the encoding process based on the feature reconstruction model, even if the account type of the target application account is an active type, only the application features corresponding to the application information of the target application account are encoded, and the information features corresponding to the information of the target application account are not encoded.

[0299] Step 680: Performing clustering processing on the reconstructed application features to obtain the clustering result of the target application account;

[0300] Exemplarily, the clustering processing is implemented by calling a clustering model; the clustering result is used to indicate at least one clustering cluster, and at least one account with the same account features is included in one clustering cluster. Exemplarily, the clustering result is obtained based on the reconstructed application features, and the clustering result is at least used to indicate the preferences shown by the target application account when installing and uninstalling applications.

[0301] It should be noted that the preferences shown when installing and uninstalling applications indicated by the clustering result can be specific preferences. For example: the accounts in the first clustering cluster all have a preference for installing network instant messaging applications, and the accounts in the second clustering cluster all have a preference for installing online collaborative office applications. The preferences indicated by the clustering result can also be abstract preferences. For example, it is only used to indicate that the accounts in the third clustering cluster have the same preferences, but does not directly indicate what kind of preferences.

[0302] Step 690: Based on the information of the fifth account, determining the information recommendation information of the fourth account that belongs to the same clustering result as the fifth account;

[0303] Exemplarily, the information of the fifth account includes the information browsing information of the fifth account in the target application. The information recommendation information is the recommendation information provided for the fourth account in the target application. The target application is usually the application program that provides the feature reconstruction model, or the host program of the application program that provides the feature reconstruction model, but it does not exclude other application programs installed and / or logged in by the fourth account.

[0304] Exemplarily, the display methods of the information recommendation information include but are not limited to at least one of the following: multimedia information stream, text information stream, push information, card information, window information.

[0305] In one example, the clustering result of the fourth account indicates a clustering cluster to which the fourth account belongs. According to the information of the fifth account with an active account type in the clustering cluster, the information recommendation information of the fourth account is determined.

[0306] The account type of the fifth account is an active type. The fifth account corresponds to information, and the information of the fifth account is the information browsing information of the fifth account in the target application.

[0307] In summary, the method provided in this embodiment reconstructs the application features through the feature reconstruction model and performs clustering processing on the reconstructed application features, ensuring the effect of clustering the accounts and avoiding the difficult clustering problem caused by the information differences between accounts of different types; determines the information recommendation information through the account clustering result, and realizes determining the information recommendation information according to the preferences of the accounts in the same clustering cluster.

[0308] Next, a detailed introduction to determining the information recommendation information of the third account is given, that is, the step 690 above has at least the following implementation manners:

[0309] Figure 12 The flowchart of the information recommendation method provided by an exemplary embodiment of the present application is shown. This method can be executed by a computer device. That is, in Figure 11 the embodiment shown, step 690 can be implemented as step 692:

[0310] Step 692: Determine the account features of the target clustering cluster according to the information of the fifth account in the target time period, and determine the information recommendation information of the fourth account according to the account features;

[0311] The target clustering cluster includes the clustering cluster indicated by the clustering result of the fourth account. Both the fourth account and the fifth account belong to the target clustering cluster. Exemplarily, the account features of the target clustering cluster include the account group portrait of the target clustering cluster. It can be understood that the fifth account is usually indicated by the clustering result obtained through clustering processing to belong to the target clustering cluster, but it does not exclude the case of being determined to belong to the target clustering cluster by other means.

[0312] Exemplarily, in this embodiment, the account type of the fourth account is an inactive type, the account type of the fifth account is an active type, and the information includes the click-through rate and click volume distribution of the fifth account for information in different categories. It can be understood that the information recommendation information of the inactive type account in the target clustering cluster is determined according to the information of the active type account in the target clustering cluster.

[0313] In an alternative implementation manner, after step 592, the following steps are further included:

[0314] Call the sorting model to sort the account features to obtain the sorting information of the information recommendation;

[0315] Add the account features of the target clustering cluster, that is, add the account group portrait of the target clustering cluster, and obtain the sorting information of the information recommendation through the sorting model. In one implementation, the account type of the fourth account corresponding to the information recommendation belongs to the inactive account.

[0316] Exemplarily, Figure 13 Shows the interface diagram of the recommendation information provided by an exemplary embodiment of the present application. In the interface diagram 450, according to the account features of the target clustering cluster, the first information 452 with the title: Weather Forecast of City X is determined as the first-order recommendation information, the second information 454 with the title: Y Community Celebrates the Festival is determined as the second-order recommendation information, and the third information 456 with the title: Weekend Promotion at Z Square is determined as the third-order recommendation information.

[0317] In summary, the method provided in this embodiment constructs an account group portrait for the accounts of the target clustering cluster by determining the account features, and obtains the information recommendation for the inactive type accounts in the same clustering cluster according to the account features, improving the information recommendation effect for the accounts of the target clustering cluster.

[0318] Figure 14 Shows the flowchart of the information recommendation method provided by an exemplary embodiment of the present application. This method can be executed by a computer device. That is, in Figure 11 In the shown embodiment, step 690 can be implemented as step 694:

[0319] Step 694: Determine the information recommendation of the fourth account according to the high-frequency information of the fifth account exceeding the target threshold within the target time period;

[0320] The target clustering cluster includes the clustering cluster indicated by the clustering result of the fourth account. Both the fourth account and the fifth account belong to the target clustering cluster. Exemplarily, in this embodiment, the account type of the fourth account is the inactive type, and the account type of the fifth account is the active type.

[0321] Exemplarily, the high-frequency information is the information whose click-through rate and / or click volume exceed the click threshold when the fourth account uses the target application within the target time period. The high-frequency auxiliary information is recalled as a candidate set as the information recommendation of the fourth account.

[0322] It can be understood that the information recommendation of the inactive type accounts in the target clustering cluster is determined according to the high-frequency information of the active type accounts in the target clustering cluster.

[0323] In summary, the method provided in this embodiment updates the recall candidate set of inactive type accounts by acquiring high-frequency information, thereby ensuring the information recommendation effect of inactive type accounts in the target cluster.

[0324] Figure 15 A schematic diagram of an account clustering model provided by an exemplary embodiment of the present application is shown.

[0325] The account clustering model 730 is trained through the first application information 712, the second application information 722 and the news information 724 to obtain a trained account clustering model 730a; illustratively, the account clustering model 730 includes a feature reconstruction model and a clustering model.

[0326] By calling the trained account clustering model 730 a , the first application information 712 and the second application information 722 are clustered to obtain a clustering result 740 of a first account corresponding to the first application information 712 and a second account corresponding to the second application information 722 .

[0327] An account group portrait 752 is generated based on the clustering result 740 of the first account and the second account, and the account group portrait 752 is added to the sorting model 754 to obtain the order information of the recommended information of the first account that is an inactive account in the clustering result 740, thereby improving the information recommendation effect of the sorting model 754 for the inactive account.

[0328] The clustering result 740 of the first account and the second account is added to the recall model 756, and the information with a click rate and a click volume exceeding the click threshold within the target time period is recalled as a candidate set, thereby improving the effect of recalling recommendations for inactive accounts.

[0329] Those skilled in the art will appreciate that the above embodiments may be implemented independently, or the above embodiments may be freely combined to form new embodiments to implement the method and / or method of using the feature reconstruction model of the present application.

[0330] Figure 16 A block diagram of a training device for a feature reconstruction model provided by an exemplary embodiment of the present application is shown. The device comprises:

[0331] The acquisition module 810 is used to acquire a first general feature of a first account, and acquire a second general feature and an auxiliary feature of a second account, wherein the first general feature includes a feature representation of first general information of the first account, the second general feature includes a feature representation of second general information of the second account, and the auxiliary feature includes a feature representation of auxiliary information of the second account; the account type of the first account is a first type, the account type of the second account is a second type, and there is no corresponding auxiliary information for an account in the first type;

[0332] An encoding module 820, configured to perform encoding processing on the first general feature to obtain a first reconstructed general feature, perform encoding processing on the second general feature to obtain a second reconstructed general feature, and perform encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature;

[0333] A prediction module 830, configured to predict the account type to which the first reconstructed general feature and the second reconstructed general feature belong, and obtain a predicted classification result;

[0334] A training module 840, configured to train the feature reconstruction model according to the contrast error and the adversarial error to obtain a trained feature reconstruction model, where the contrast error includes the error between the second reconstructed general feature and the reconstructed auxiliary feature, and the adversarial error includes the error between the predicted classification result and the actual classification result, and the actual classification result is the account type actually corresponding to the first general feature and the second general feature.

[0335] In an alternative design of this embodiment, the feature reconstruction model includes a contrastive learning encoder;

[0336] The encoding module 820 is further configured to:

[0337] Call the contrastive learning encoder to perform encoding processing on the first general feature to obtain the first reconstructed general feature, call the contrastive learning encoder to perform encoding processing on the second general feature to obtain the second reconstructed general feature, and call the contrastive learning encoder to perform encoding processing on the auxiliary feature to obtain the reconstructed auxiliary feature.

[0338] In an alternative design of this embodiment, the feature reconstruction model further includes a general encoder and an auxiliary encoder;

[0339] The obtaining module 810 is further configured to:

[0340] Obtain the first general information of the first account, and obtain the second general information and the auxiliary information of the second account;

[0341] Call the general encoder to perform encoding processing on the first general information to obtain the first general feature, call the general encoder to perform encoding processing on the second general information to obtain the second general feature, and call the auxiliary encoder to perform encoding processing on the auxiliary information to obtain the auxiliary feature.

[0342] In an alternative design of this embodiment, the first general information and the second general information are multi-hot encoded vectors, and the auxiliary information is a continuous value vector.

[0343] In an alternative design of this embodiment, the feature reconstruction model further includes a general decoder and an auxiliary decoder;

[0344] The apparatus further includes:

[0345] A decoding module 850, configured to call the general decoder to decode the first general feature to obtain first reconstructed general information, call the general decoder to decode the second general feature to obtain second reconstructed general information, and call the auxiliary decoder to decode the auxiliary feature to obtain reconstructed auxiliary information;

[0346] The training module 840 is further configured to:

[0347] Train the feature reconstruction model according to the comparison error, the adversarial error, and the reconstruction error to obtain the trained feature reconstruction model;

[0348] Wherein, the reconstruction error includes at least one of the error between the first reconstructed general information and the first general information, the error between the second reconstructed general information and the second general information, and the error between the reconstructed auxiliary information and the auxiliary information.

[0349] In an alternative design of this embodiment, at least one of the comparison error, the adversarial error, and the reconstruction error includes a cross-entropy loss.

[0350] In an alternative design of this embodiment, the feature reconstruction model includes a gradient reversal layer and an account classifier;

[0351] The prediction module 830 is further configured to:

[0352] Call the gradient reversal layer to perform reversal processing on the first reconstructed general feature and the second reconstructed general feature to obtain a processed first reconstructed general feature and a processed second reconstructed general feature;

[0353] Call the account classifier to predict the account types of the processed first reconstructed general feature and the processed second reconstructed general feature to obtain the predicted classification result.

[0354] In an alternative design of this embodiment, the first general information is first application information, the first application information includes installation and uninstallation information of at least one application by the first account, the second general information is second application information, the second application information includes installation and uninstallation information of at least one application by the second account, the auxiliary information is information, and the information includes information browsing information of the second account in the target application;

[0355] The obtaining module 810 is further configured to:

[0356] Obtain a first application feature of the first account, and obtain a second application feature and an information feature of the second account, where the first application feature includes a feature representation of the first application information of the first account, the second application feature includes a feature representation of the second application information of the second account, and the information feature includes a feature representation of the information of the second account;

[0357] The encoding module 820 is further configured to:

[0358] Perform encoding processing on the first application feature to obtain a first reconstructed application feature, perform encoding processing on the second application feature to obtain a second reconstructed application feature, and perform encoding processing on the information feature to obtain a reconstructed information feature;

[0359] The prediction module 830 is further configured to:

[0360] Predict the account types to which the first reconstructed application feature and the second reconstructed application feature belong, and obtain the predicted classification result.

[0361] Figure 17 The block diagram of a device for using a feature reconstruction model provided by an exemplary embodiment of the present application is shown. The device includes:

[0362] An obtaining module 860, configured to obtain a general feature of at least one third account, where the general feature includes a feature representation of the general information of the third account;

[0363] An encoding module 870, configured to perform encoding processing on the general feature based on the feature reconstruction model to obtain a reconstructed general feature;

[0364] A clustering module 880, configured to perform clustering processing on the reconstructed general feature to obtain a clustering result of the third account.

[0365] In an alternative design of this embodiment, the device further includes:

[0366] A determination module 890, configured to obtain auxiliary information of the third account within a target time period, and determine account features of a target clustering cluster according to the auxiliary information, where the target clustering cluster includes the clustering cluster indicated by the clustering result of the third account.

[0367] In an alternative design of this embodiment, the determination module 890 is further configured to:

[0368] Call a sorting model to perform sorting processing on the account features to obtain recommendation information of a to-be-recommended account, where the to-be-recommended account belongs to the target clustering cluster.

[0369] In an alternative design of this embodiment, the device further includes:

[0370] A determination module 890, configured to obtain high-frequency auxiliary information of the third account that exceeds a target threshold within a target time period, and determine recommendation information of a to-be-recommended account according to the high-frequency auxiliary information. The third account and the to-be-recommended account belong to a target clustering cluster, and the target clustering cluster includes the clustering cluster indicated by the clustering result of the third account.

[0371] Figure 18 The block diagram of an information recommendation device provided by an exemplary embodiment of the present application is shown. The device includes:

[0372] An acquisition module 910, configured to acquire application characteristics of multiple target application accounts. The target application accounts include at least one fourth account of an inactive account type and at least one fifth account of an active account type. The application characteristics include a feature representation of application information of the target application accounts, and the application information includes installation and uninstallation information of the target application accounts for at least one application;

[0373] An encoding module 920, configured to perform encoding processing on the application characteristics based on a feature reconstruction model to obtain reconstructed application characteristics;

[0374] A clustering module 930, configured to perform clustering processing on the reconstructed application characteristics to obtain a clustering result of the target application accounts;

[0375] A determination module 940, configured to determine information recommendation information of the fourth account that belongs to the same clustering result as the fifth account based on the information of the fifth account. The information includes information browsing information of the fifth account in a target application.

[0376] It should be noted that when the device provided in the above embodiment implements its functions, only the division of the above-mentioned respective function modules is used for illustration. In actual applications, the above functions may be allocated to different function modules according to actual needs, that is, the content structure of the device is divided into different function modules to complete all or part of the functions described above.

[0377] Regarding the device in the above embodiment, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method; the technical effects obtained by each module performing operations are the same as those in the embodiments related to the method, and will not be elaborated in detail here.

[0378] The embodiments of the present application also provide a computer device, which includes: a processor and a memory, and a computer program is stored in the memory; the processor is configured to execute the computer program in the memory to implement the method for reconstructing a feature model and / or the usage method provided in each of the above method embodiments.

[0379] Optionally, the computer device is a server. Exemplarily, Figure 19 It is a structural block diagram of a server provided by an exemplary embodiment of the present application.

[0380] Generally, the server 2300 includes: a processor 2301 and a memory 2302.

[0381] The processor 2301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 2301 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 2301 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the wake state, also known as the central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 2301 may be integrated with a graphics processing unit (GPU), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 2301 may further include an artificial intelligence (AI) processor, and the AI processor is used to process computational operations related to machine learning.

[0382] The memory 2302 may include one or more computer-readable storage media, and the computer-readable storage media may be non-transitory. The memory 2302 may further include a high-speed random access memory and a non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 2302 is used to store at least one instruction, and the at least one instruction is used to be executed by the processor 2301 to implement the method for reconstructing a feature model and / or the usage method provided in the method embodiments of the present application.

[0383] In some embodiments, the server 2300 may further optionally include: an input interface 2303 and an output interface 2304. The processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 may be connected through a bus or signal lines. Each peripheral device may be connected to the input interface 2303 and the output interface 2304 through a bus, signal lines, or a circuit board. The input interface 2303 and the output interface 2304 may be used to connect at least one peripheral device related to input / output (I / O) to the processor 2301 and the memory 2302. In some embodiments, the processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 2301, the memory 2302, the input interface 2303, and the output interface 2304 may be implemented on a separate chip or circuit board, and the embodiments of the present application do not limit this.

[0384] Those skilled in the art can understand that the structures shown above do not constitute a limitation on the server 2300, and it may include more or fewer components than shown in the figure, or combine certain components, or adopt different component arrangements.

[0385] In an exemplary embodiment, a chip is further provided. The chip includes programmable logic circuits and / or program instructions, and when the chip runs on a computer device, it is used to implement the method and / or usage method and / or information recommendation method of the feature reconstruction model described in the above aspects.

[0386] In an exemplary embodiment, a computer program product is further provided. The computer program product includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor reads and executes the computer instructions from the computer-readable storage medium to implement the method and / or usage method and / or information recommendation method of the feature reconstruction model provided in the above method embodiments.

[0387] In an exemplary embodiment, a computer-readable storage medium is further provided. The computer-readable storage medium stores a computer program, and the computer program is loaded and executed by a processor to implement the method and / or usage method and / or information recommendation method of the feature reconstruction model provided in the above method embodiments.

[0388] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium, and the storage medium mentioned above can be a read-only memory, a magnetic disk, or an optical disc, etc.

[0389] Those skilled in the art should be able to realize that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented by hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. The computer-readable medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transmission of a computer program from one place to another. The storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0390] The above are only alternative embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A training method for a feature reconstruction model, characterized in that, The method includes: Obtaining a first general feature of a first account, and obtaining a second general feature and an auxiliary feature of a second account. The first general feature includes a feature representation of first general information of the first account, where the first general information is information determined or obtained when registering the first account. The second general feature includes a feature representation of second general information of the second account, where the second general information is information determined or obtained when registering the second account. The auxiliary feature includes a feature representation of auxiliary information of the second account, where the auxiliary information is information generated during the application process of the second account. The account type of the first account is a first type, and the account type of the second account is a second type. There is no corresponding auxiliary information for accounts of the first type. Performing encoding processing on the first general feature to obtain a first reconstructed general feature, performing encoding processing on the second general feature to obtain a second reconstructed general feature, and performing encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature. Predicting the account types to which the first reconstructed general feature and the second reconstructed general feature belong to obtain a predicted classification result. Training the feature reconstruction model according to a contrast error and an adversarial error to obtain a trained feature reconstruction model. The contrast error includes the error between the second reconstructed general feature and the reconstructed auxiliary feature, and the adversarial error includes the error between the predicted classification result and the actual classification result. The actual classification result is the actual corresponding account types of the first general feature and the second general feature.

2. The method according to claim 1, wherein The feature reconstruction model includes a contrastive learning encoder. The performing encoding processing on the first general feature to obtain a first reconstructed general feature, performing encoding processing on the second general feature to obtain a second reconstructed general feature, and performing encoding processing on the auxiliary feature to obtain a reconstructed auxiliary feature includes: Invoking the contrastive learning encoder to perform encoding processing on the first general feature to obtain the first reconstructed general feature, invoking the contrastive learning encoder to perform encoding processing on the second general feature to obtain the second reconstructed general feature, and invoking the contrastive learning encoder to perform encoding processing on the auxiliary feature to obtain the reconstructed auxiliary feature.

3. The method according to claim 2, characterized in that, The feature reconstruction model further includes a general encoder and an auxiliary encoder. The obtaining a first general feature of a first account, and obtaining a second general feature and an auxiliary feature of a second account includes: Obtaining the first general information of the first account, and obtaining the second general information and the auxiliary information of the second account. Invoking the general encoder to perform encoding processing on the first general information to obtain the first general feature, invoking the general encoder to perform encoding processing on the second general information to obtain the second general feature, and invoking the auxiliary encoder to perform encoding processing on the auxiliary information to obtain the auxiliary feature.

4. The method according to claim 3, wherein The feature reconstruction model further includes a general decoder and an auxiliary decoder. The method further includes: Call the general decoder to decode the first general feature to obtain the first reconstructed general information, call the general decoder to decode the second general feature to obtain the second reconstructed general information, and call the auxiliary decoder to decode the auxiliary feature to obtain the reconstructed auxiliary information; Training the feature reconstruction model according to the contrast error and the adversarial error to obtain the trained feature reconstruction model includes: Training the feature reconstruction model according to the contrast error, the adversarial error, and the reconstruction error to obtain the trained feature reconstruction model; Wherein, the reconstruction error includes at least one of the error between the first reconstructed general information and the first general information, the error between the second reconstructed general information and the second general information, and the error between the reconstructed auxiliary information and the auxiliary information.

5. The method according to any one of claims 1 to 4, characterized in that The feature reconstruction model includes a gradient reversal layer and an account classifier; Predicting the account types to which the first reconstructed general feature and the second reconstructed general feature belong to obtain a predicted classification result, including: Calling the gradient reversal layer to perform reversal processing on the first reconstructed general feature and the second reconstructed general feature to obtain the processed first reconstructed general feature and the processed second reconstructed general feature; Calling the account classifier to predict the account types of the processed first reconstructed general feature and the processed second reconstructed general feature to obtain the predicted classification result.

6. The method according to claim 1, characterized in that, The first general information is the first application information, the first application information includes the installation and uninstallation information of at least one application by the first account, the second general information is the second application information, the second application information includes the installation and uninstallation information of at least one application by the second account, the auxiliary information is the information, and the information includes the information browsing information of the second account in the target application; Obtaining the first general feature of the first account, and obtaining the second general feature and the auxiliary feature of the second account, including: Obtaining the first application feature of the first account, and obtaining the second application feature and the information feature of the second account, the first application feature includes the feature representation of the first application information of the first account, the second application feature includes the feature representation of the second application information of the second account, and the information feature includes the feature representation of the information of the second account; Encoding the first general feature to obtain the first reconstructed general feature, encoding the second general feature to obtain the second reconstructed general feature, and encoding the auxiliary feature to obtain the reconstructed auxiliary feature, including: Encoding the first application feature to obtain the first reconstructed application feature, encoding the second application feature to obtain the second reconstructed application feature, and encoding the information feature to obtain the reconstructed information feature; Predicting the account types to which the first reconstructed general feature and the second reconstructed general feature belong to obtain a predicted classification result, including: Predict the account type to which the first reconstructed application feature and the second reconstructed application feature belong to obtain the predicted classification result.

7. A method for using a feature reconstruction model, characterized in that, The feature reconstruction model is trained by the method according to any one of claims 1 to 6; The method includes: Obtain general features of at least one third account, where the general features include feature representations of general information of the third account; Perform encoding processing on the general features based on the feature reconstruction model to obtain reconstructed general features; Perform clustering processing on the reconstructed general features to obtain the clustering result of the third account.

8. An information recommendation method, characterized in that, The method includes: Obtain application features of multiple target application accounts, where the target application accounts include at least one fourth account with an account type of inactive type and at least one fifth account with an account type of active type, the application features include feature representations of application information of the target application accounts, and the application information includes installation and uninstallation information of the target application accounts for at least one application; Perform encoding processing on the application features based on the feature reconstruction model to obtain reconstructed application features, where the reconstruction model is trained by the method according to any one of claims 1 to 6; Perform clustering processing on the reconstructed application features to obtain the clustering result of the target application accounts; Based on the information of the fifth account, determine the information recommendation information of the fourth account that belongs to the same clustering result as the fifth account, where the information includes information browsing information of the fifth account in the target application.

9. A training device for a feature reconstruction model, characterized in that, The device includes: An acquisition module, configured to acquire the first general features of the first account, and acquire the second general features and auxiliary features of the second account. The first general features include feature representations of the first general information of the first account, and the first general information is information determined or acquired when registering the first account. The second general features include feature representations of the second general information of the second account, and the second general information is information determined or acquired when registering the second account. The auxiliary features include feature representations of the auxiliary information of the second account, and the auxiliary information is information generated during the application process of the second account. The account type of the first account is the first type, and the account type of the second account is the second type, and there is no corresponding auxiliary information for the accounts of the first type; An encoding module, configured to perform encoding processing on the first general features to obtain first reconstructed general features, perform encoding processing on the second general features to obtain second reconstructed general features, and perform encoding processing on the auxiliary features to obtain reconstructed auxiliary features; A prediction module, configured to predict the account type to which the first reconstructed general features and the second reconstructed general features belong to obtain a predicted classification result; A training module, configured to train the feature reconstruction model according to a contrast error and an adversarial error to obtain a trained feature reconstruction model, where the contrast error includes an error between the second reconstructed general feature and the reconstructed auxiliary feature, and the adversarial error includes an error between the predicted classification result and the actual classification result, and the actual classification result is the account type actually corresponding to the first general feature and the second general feature.

10. An apparatus for using a feature reconstruction model, characterized in that, The feature reconstruction model is trained by the method according to any one of claims 1 to 6; the apparatus includes: An acquisition module, configured to acquire general features of at least one third account, where the general features include feature representations of general information of the third account; An encoding module, configured to perform encoding processing on the general features based on the feature reconstruction model to obtain reconstructed general features; A clustering module, configured to perform clustering processing on the reconstructed general features to obtain a clustering result of the third account.

11. An information recommendation device, characterized in that, The apparatus includes: An acquisition module, configured to acquire application features of multiple target application accounts, where the target application accounts include at least one fourth account with an account type of inactive type and at least one fifth account with an account type of active type, the application features include feature representations of application information of the target application accounts, and the application information includes installation and uninstallation information of the target application accounts for at least one application; An encoding module, configured to perform encoding processing on the application features based on a feature reconstruction model to obtain reconstructed application features, where the reconstruction model is trained by the method according to any one of claims 1 to 6; A clustering module, configured to perform clustering processing on the reconstructed application features to obtain a clustering result of the target application accounts; A determination module, configured to determine information recommendation information of the fourth account belonging to the same clustering result as the fifth account based on the information of the fifth account, where the information includes information browsing information of the fifth account in the target application.

12. A computer device, characterized in that, The computer device includes: a processor and a memory, where at least one program is stored in the memory; the processor is configured to execute the at least one program in the memory to implement the training method of the feature reconstruction model according to any one of claims 1 to 6, or the using method of the feature reconstruction model according to claim 7, or the information recommendation method according to claim 8.

13. A computer-readable storage medium, characterized in that, An executable instruction is stored in the readable storage medium, and the executable instruction is loaded and executed by a processor to implement the training method of the feature reconstruction model according to any one of claims 1 to 6, or the using method of the feature reconstruction model according to claim 7, or the information recommendation method according to claim 8.

14. A computer program product, characterized in that, The computer program product includes computer instructions, where the computer instructions are stored in a computer-readable storage medium, and the processor reads and executes the computer instructions from the computer-readable storage medium to implement the training method of the feature reconstruction model according to any one of claims 1 to 6, or the using method of the feature reconstruction model according to claim 7, or the information recommendation method according to claim 8.

Images