CN112749332B

CN112749332B - Data processing method, device and computer readable medium

Info

Publication number: CN112749332B
Application number: CN202010662408.0A
Authority: CN
Inventors: 郝晓波; 刘雨丹; 葛凯凯; 唐琳瑶; 张旭; 林乐宇
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-07-10
Filing date: 2020-07-10
Publication date: 2023-12-15
Anticipated expiration: 2040-07-10
Also published as: CN112749332A

Abstract

The application discloses a data processing method, a data processing device and a computer readable medium, and relates to the technical field of computers. The method comprises the following steps: acquiring a first score determined by an interest heuristic model to be trained according to user characteristic data, wherein the first score is used for representing the possibility that sample data belong to unknown interests of a user; acquiring a weight value determined by the joint probability model according to sample data, wherein the weight value is used for representing the possibility that a sample is suitable for training an interest heuristic model; the first score is adjusted according to the weight value, and the weight value is positively correlated with the first score; and training an interest heuristic model to be trained according to the adjusted first score, wherein the trained interest heuristic model is used for determining estimated interest information of the user according to the user characteristic data. Therefore, determining the content to be pushed according to the estimated interest information of the user; the content to be pushed is pushed to the client corresponding to the user, so that the content can be pushed to the user according to the unknown interests of the user, and the diversity of the pushed content is improved.

Description

Data processing method, device and computer readable medium

Technical Field

The present application relates to the field of computer technology, and more particularly, to a data processing method, apparatus, and computer readable medium.

Background

With the continuous development of internet applications, the time for users to use the internet is more and more, and how to enable users to quickly obtain possibly needed information in massive information, so that user experience is further improved, and information pushing services are needed to be provided for users. At present, when pushing content, a new content is often pushed for a user according to high-frequency clicks of some historical content or personal interests pre-entered by the user, however, the pushing manner can cause that the pushed content includes excessive repeated content, so that the pushed content is single.

Disclosure of Invention

The present application proposes a data processing method, apparatus and computer readable medium to ameliorate the above disadvantages.

In a first aspect, an embodiment of the present application provides a data processing method, including: acquiring a first score determined by an interest heuristic model to be trained according to user characteristic data, wherein the first score is used for representing the possibility that sample data belong to unknown interests of a user; acquiring a weight value determined by the joint probability model according to the sample data, wherein the weight value is used for representing the possibility that the sample is suitable for training the interest heuristic model; adjusting the first score according to the weight value, wherein the weight value is positively correlated with the first score; training the interest heuristic model to be trained according to the adjusted first score, wherein the trained interest heuristic model is used for determining estimated interest information of a user according to the user characteristic data, the estimated interest information is unknown interest information of the user, and the estimated interest information is used for determining content to be pushed to a client corresponding to the user.

In a second aspect, an embodiment of the present application further provides a data processing apparatus, including: the device comprises a first acquisition unit, a second acquisition unit, an adjustment unit and a training unit. The first acquisition unit is used for acquiring a first score determined by the interest heuristic model to be trained according to the user characteristic data, wherein the first score is used for representing the possibility that the sample data belong to the unknown interests of the user. And the second acquisition unit is used for acquiring weight values determined by the joint probability model according to the sample data, wherein the weight values are used for representing the possibility that the sample is suitable for training the interest heuristic model. And the adjusting unit is used for adjusting the first score according to the weight value, and the weight value is positively correlated with the first score. The training unit is used for training the interest heuristic model to be trained according to the adjusted first score, the trained interest heuristic model is used for determining estimated interest information of a user according to the user characteristic data, the estimated interest information is unknown interest information of the user, and the estimated interest information is used for determining content to be pushed to a client corresponding to the user.

In a third aspect, embodiments of the present application also provide a computer readable storage medium storing program code executable by a processor, the program code when executed by the processor causing the processor to perform the above method.

The data processing method, the data processing device and the computer readable medium provided by the application acquire the estimated interest information of the user when the content is required to be pushed for the user. The method comprises the steps that estimated interest information is unknown interest information of a user, specifically, the estimated interest information is obtained according to a trained interest heuristic model, and the training process of the interest heuristic model is that a first score which is determined by the interest heuristic model to be trained according to user characteristic data is obtained, wherein the first score is used for representing the possibility that sample data belong to unknown interests of the user; acquiring a weight value determined by the joint probability model according to the sample data, wherein the weight value is used for representing the possibility that the sample is suitable for training the interest heuristic model; and adjusting the first score according to the weight value, wherein the weight value is positively correlated with the first score, so that the first score can be used for considering the applicability of the sample for training the interest heuristic model by correcting the first score output by the weight value for training the interest heuristic model to be trained, the training of the interest heuristic model is more reasonable, and the obtained estimated interest information is more accurate. Then, determining content to be pushed according to the estimated interest information of the user; the content to be pushed is pushed to the client corresponding to the user, so that the content pushing method and the content pushing device not only can push the content for the user according to the known interests of the user, but also can push the content for the user according to the unknown interests of the user, and the diversity of the pushed content is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 shows a schematic diagram of an information pushing system according to an embodiment of the present application;

FIG. 2 is a flow chart of a method for data processing according to an embodiment of the present application;

FIG. 3 is a flow chart of a method for data processing according to another embodiment of the present application;

FIG. 4 is a schematic structural diagram of a joint probability model provided by an embodiment of the present application;

FIG. 5 is a flow chart of a method for data processing according to another embodiment of the present application;

FIG. 6 is a schematic diagram of an interest heuristic model provided by an embodiment of the present application;

FIG. 7 is a schematic diagram of a joint training model according to an embodiment of the present application;

FIG. 8 is a flow chart of a method for data processing according to still another embodiment of the present application;

FIG. 9 is a schematic diagram of a client-side assignment interface according to an embodiment of the present application;

FIG. 10 is a schematic diagram of a detail interface of content of a client according to an embodiment of the present application;

FIG. 11 is a schematic diagram of a client-side assignment interface according to another embodiment of the present application;

FIG. 12 is a schematic diagram of a client-side assignment interface provided in accordance with yet another embodiment of the present application;

FIG. 13 is a flow chart of a method for data processing according to still another embodiment of the present application;

FIG. 14 is a schematic view of a scenario in which an interest heuristic model provided in an embodiment of the present application is applied to information push;

FIG. 15 is a schematic diagram of a client-side assignment interface according to yet another embodiment of the present application;

FIG. 16 is a block diagram of a data processing apparatus according to an embodiment of the present application;

FIG. 17 shows a block diagram of an electronic device provided by an embodiment of the application;

fig. 18 shows a storage unit for storing or carrying program codes for implementing the information push method according to the embodiment of the present application.

Detailed Description

In order to enable those skilled in the art to better understand the present application, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present application with reference to the accompanying drawings.

With the continuous development of internet applications, the time for users to use the internet is more and more, and how to enable users to quickly obtain possibly needed information in massive information, so that user experience is further improved, and information pushing services are needed to be provided for users.

Referring to fig. 1, fig. 1 shows an information push system provided by an embodiment of the present application. As shown in fig. 1, the information push system includes a server 100 and a user terminal 200. The server 100 and the user terminal 200 are located in a wireless network or a wired network, and data interaction between the server 100 and the user terminal 200 is enabled.

In some embodiments, when the user logs in at the user terminal through the account, all information corresponding to the account may be stored in the storage space of the server 100. The server 100 may be a separate server, may be a server cluster, may be a local server, or may be a cloud server.

The user terminal has a plurality of applications installed therein and the server 100 can push some contents to the user terminal, specifically, may be a certain application that pushes the content to the user terminal, and the application displays the content, so that the content can be pushed to the user corresponding to the user terminal.

The server 100 may be connected to a plurality of user terminals, and may push content to be pushed to all user terminals, or may select one of the user terminals according to some policies, and push content to be pushed to the selected user terminal. And the specific policy may be determined according to the content to be pushed and the users corresponding to the respective user terminals. In the embodiment of the present application, the content to be pushed may be information such as an article, a video, and a picture.

However, the inventors found in the study that the push mode of the current content is less effective.

In an information recommendation system, a traditional recommendation algorithm makes recommendations based on historical behaviors and semantic features of users and contents, and takes the click rate and the click frequency of the users as measurement targets. However, determining the content to be pushed according to the click of the user may result in repeated recommendation of the high-frequency click interest and the known interest of the user, resulting in convergence of the field of view of the user, and the recommendation system may not meet all interests of the user, resulting in degradation of user experience and even loss of the user. This is detrimental to the diversity of the recommendation system and the long-term development of the recommendation system.

For example, a click rate estimation method is used for pushing content for users, specifically, for each pair of combination of the users and the content, the click probability of the users on the content is predicted by using a deep neural network model according to basic characteristics, behavior characteristics, content and contextual characteristics of the users. For example, the model may include a support decomposer neural network (FNN), a Product-based neural network (Product-based Neural Network, PNN), a generalized linear Deep neural network (Wide Linear Model and Deep Netural Network, wide & Deep), a Deep decomposer (Deep Factorization Machines, deep FM), and the like. The network structure of the model is not greatly different, mainly the difference of calculation modes of some features is that firstly high-dimensional sparse features are embedded into low-dimensional continuous vectors, then co-occurrence relations and nonlinear changes among the features are learned through all calculation layers of the model, and finally the estimated click rate fraction of the interval in [0,1] is output.

As another example, content is pushed to users in a exploration and utilization method, and in particular, the exploration and utilization method is a method that is more widely used by a recommender system to explore user interests. . The method treats the known interests of the user as the existing benefits, treats the unknown interests of the user as the explorable benefits, and selects to utilize the currently known benefits or explore some new strategies to promote future benefits according to the existing experience at the time of decision. The algorithm of the method mainly comprises confidence interval upper bound (Upper Confidence Bound, UCB), thompson (Thompson), linear UCB (LinUCB) and other algorithms. The main idea is to balance the exploration and exploitation degree of interests by a multi-arm slot machine method, and update the decision benefit of the user by click feedback of the user so as to achieve a globally optimal benefit.

For another example, content is pushed to a user by a Look-alike method, specifically, the method firstly determines the user clicking the content to be recommended as a seed user, and then calculates a user similar to the seed user as a recommended target user through a Look-alike model. Common Look-alike models include similarity-based models and regression-based models. And after the user data is subjected to feature vectorization based on a similarity model such as local sensitive hash (Locality Sensitive Hashing, LSH), cosine similarity and a mixed model, the similarity of the seed user and the candidate user is measured by calculating the distance between vectors. Regression-based models such as logistic regression (logistics regression, LR) and gradient lifting iterative decision tree (Gradient Boosting Decision Tree, GBDT) are used for training a regression model for each user or content, taking a seed user as a positive sample of a corresponding feature, randomly extracting non-seed users as negative samples of the model, and taking the score of the model output interval in [0,1] as the click probability of the candidate user on the content.

For click rate estimation and look-alike methods, the model relies on portraits of known interests of the user and is not targeted based on click rate and number of clicks, which can result in the recommendation system over-utilizing the characterization of the known interests of the user, recommending too many of the same interest content to the user. The user is not interested by the recommendation system, and the recommendation system is difficult to enrich the interest of the user because of high recommendation cost and low click rate of the user, and the recommended content visual field is limited.

For the exploration and utilization method, the model directly models the long-term benefits of the interests of the user, and considers the known interest portrait benefits and the unknown interest portrait benefits of the user, so that the requirements are met to a certain extent. However, the existing exploration and utilization methods still have some problems: firstly, the existing model has simple characteristics, and complicated characteristic calculation is difficult to introduce; in addition, the model is based on the strong assumption of probability distribution, and the probability distribution is judged through a confidence interval; finally, the model distinguishes the boundary between exploitation and exploration by setting parameters, which is an effective boundary which cannot be distinguished accurately between exploitation and exploration, and can influence the judgment of the model on the true global optimal profit, so that the pushing result is inaccurate.

Accordingly, to overcome the above-described deficiencies, embodiments of the present application provide a data processing method that is capable of determining an unknown interest of a user in order to push content to the user based on the unknown interest of the user. As shown in fig. 2, in particular, the execution subject of the method may be the server described above, and the method includes: s201 to S204.

S201: and obtaining a first score determined by the interest heuristic model to be trained according to the user characteristic data.

Wherein the first score is used to characterize a likelihood that the sample data belongs to an unknown interest of the user, i.e., a first likelihood.

The interest heuristic model can output interest information of a user according to the similarity between the user characteristic data and a preset interest label, and the unknown interest output by the interest heuristic model can be more accurate along with continuous model training and learning.

In one embodiment, the user characteristic data includes behavior data generated by a user for a specified application module of a client and operations of other application modules associated with the specified application module and belonging to the client. Specifically, the client includes a plurality of application modules, each application module corresponding to one functional service of the client. For example, the client may be a game client, and each game scene in the game client corresponds to one application module, for example, the clearance mode and the player engagement mode each correspond to one application module, that is, different application modules provide different functional services. For another example, the client is a social client, for example, the client is a WeChat APP, and the application modules corresponding to the client may include a "watch at once" module, a friend circle module, a chat module, a public number module, and an applet module.

In one embodiment, the other application modules related to the specified application module may be application modules having the same or similar functions as the specified application module, and in the embodiment of the present application, the specified application module has a content pushing function, and the other application modules related to the specified application module also have a content pushing function.

In the embodiment of the present application, the designated application module is a "see-through" module, and then other application modules related to the designated application module may include public numbers, applets, and the like. The behavior data of the user can reflect the operation data of the user on the content pushed by the appointed application module and other application modules related to the appointed application module, for example, the operation data comprises the clicking times, the clicking frequency and the like, and the preference of the user can be reflected to a certain extent.

As an embodiment, the interest heuristic model to be trained determines, from the user feature data, a likelihood that the sample data belongs to an unknown interest of the user, i.e. a first likelihood, and the accuracy of the first likelihood of its determination is higher as the interest heuristic model is trained. Specifically, the interest heuristic model can obtain estimated interest information of the user according to the user characteristic data. It should be noted that, when the interest heuristic model is not trained, that is, the accuracy of the estimated interest information output by the initial interest heuristic model is poor, the accuracy of the estimated interest information output by the interest heuristic model needs to be improved in a training manner.

Specifically, the initial unknown interest information determined by the interest heuristic model to be trained according to the user feature data may be an interest tag, and then the interest heuristic model is matched with the sample data, for example, a keyword of the sample data is matched with the interest tag, and whether the sample belongs to the unknown interest information of the user is determined according to the matching result.

S202: and acquiring a weight value determined by the joint probability model according to the sample data.

The weight value is used to characterize the likelihood that the sample is suitable for training the heuristic model of interest.

Specifically, the sample data is exposure data corresponding to the user. Specifically, the exposure data is content processed by the user in a preset time period, wherein the processed content may include at least one of clicking, browsing and commenting content of the user, and in the embodiment of the present application, the sample data is content clicked by the user in the preset time period. However, the user click sample may be of a known interest to the user, e.g., the server determines content pushed to the user based on the known interest information of the user, and the user clicks on the content while browsing the content, and the user clicks on the content, possibly because the user clicks while reading non-pushed content, e.g., the user sees content posted by friends in a friend circle, and the content does not match the known interest information of the user. If the samples used in training the interest heuristic model to be trained are mostly content pushed to the user based on the known interest information of the user, it may result in that the unknown interest of the user obtained by the interest heuristic model after training may be excessively similar to the known interest of the user, so that a larger portion of the content pushed based on the unknown interest information of the user and the content pushed based on the known interest information of the user may be duplicated. Thus, in order to avoid training the interest heuristic model to be trained directly by clicking on the content of the user based on the known interest information of the user, a weight value is set, which may be a value of [0,1], i.e. the weight value may be a value greater than or equal to 0 and less than or equal to 1.

As an embodiment, the manner in which the weight value of the sample data is determined by the joint probability model may be empirically set, for example, when the sample data is acquired, one weight value is empirically set for each sample data.

As another implementation manner, a history push record of the user may be obtained, specifically, the server pushes content for the user based on the known interest information of the user, then the pushed content is named as interest content, and an identity is set for each interest content pushed by the user, for example, the identity may be a title or a content abstract of the pushed interest content. Then, the click of the user on each content is recorded, so that whether the content clicked by the user is interesting content or not can be determined, if the content is interesting content, a lower weight value is set for the content, and if the content is not interesting content, a higher weight value is set for the content.

As yet another implementation manner, the weight value may be determined according to social information of the user, specifically, please refer to the following examples.

S203: and adjusting the first score according to the weight value.

Specifically, the weight value can correct the first score determined by the interest heuristic model to be trained according to the user feature data, i.e. the weight value can influence the magnitude of the first score. As one embodiment, the weight value is positively correlated with the first score. The positive correlation means that the weight value and the first score increase synchronously, namely the weight value increases, the first score also increases, the weight value decreases, and the first score also decreases. That is, the larger the weight value, the larger the first score can be made, and the smaller the weight value, the smaller the first score can be made.

The first score is used to characterize the likelihood that the sample data belongs to the user's unknown interests, that is, the interest heuristic model scores a sample, i.e., the higher the first score is output, the higher the likelihood that the sample is likely to be an unknown interest to the user. The weight value characterizes the possibility that the sample is suitable for training the interest heuristic model, which reflects the applicability that the sample can be used for training the interest heuristic model, the larger the weight value is, the more suitable the sample is for training the interest heuristic model, the smaller the weight value is, the less suitable the sample is for training the interest heuristic model, so if a certain sample is not suitable for training the interest heuristic model, a lower weight value can be set for the sample.

In the method of the multi-arm slot machine, the data of a plurality of samples are definitely divided into two sample sets, namely, one sample set is a sample for training one model, the other sample set is a sample for training the other model, and the samples in the two sample sets are different, or, although some samples in the two sample sets are the same, the scoring value of the interest heuristic model to be trained on the samples only depends on the interest heuristic model itself. That is, the method of the multi-arm slot machine simply classifies a sample, e.g., if the sample is not 80% likely to be pushed by the server to the user based on the user's known interest tag, then the sample is assigned to a first class, wherein the sample within the first class is used to train the interest heuristic model, and if the sample is not 20% likely to be pushed by the server to the user based on the user's known interest tag, then the sample is assigned to a second class, wherein the sample is not used to train the interest heuristic model, i.e., the sample is discarded, i.e., the sample is not used to train the interest heuristic model.

In the embodiment of the present application, each sample data obtained can be used to train the interest heuristic model, because, even if there is a 20% likelihood that the sample is not pushed by the server to the user based on the known interest information for the sample, the sample can reflect the interest of the user to some extent, because, after all, it is the content clicked on by the user, it should still be used to train the interest heuristic model, and in order to avoid that the sample may belong to the known interest for which the sample is used, if the first score of the interest heuristic model for the sample is too high, there may be a certain unknown interest too similar to the known interest, so the score determined for the sample by the interest heuristic model can be corrected by means of weights. Thus, in case the likelihood that the sample belongs to the content pushed by the server for the user based on the known interest information for the user is low, a low weight value may be output to correct the score of the interest heuristic model, i.e. the first score is reduced, so that the likelihood that the interest heuristic model considers that the sample belongs to the non-performed interest of the user is low, i.e. the likelihood that the key features in the sample can act as unknown interest tags of the user is low. The key feature may be the type of sample, title, keyword or entry, etc.

S204: and training the interest heuristic model to be trained according to the adjusted first score.

The trained interest heuristic model is used for determining estimated interest information of a user according to the user characteristic data, the estimated interest information is unknown interest information of the user, and the estimated interest information is used for determining content to be pushed to a client corresponding to the user.

The interest heuristic model can determine the possibility that the sample data is the unknown interest of the user according to the currently determined unknown interest information, namely, the first possibility, the output first possibility can be regarded as a predicted value of the interest heuristic model on the sample data, specifically, the user characteristic data is taken as an input value of the interest heuristic model, the interest heuristic model can determine the unknown interest information of the user according to the user characteristic data, wherein the non-executed interest information can be an unexecuted interest label of the user, then the unknown interest information of the user is matched with the sample, the matching degree of the unknown interest information of the user and the sample is calculated, so that the possibility that the sample belongs to the unknown interest of the user is obtained, namely, the first score can be taken as a transmission value of a loss function of the interest heuristic model, the loss function is continuously optimized through the transmission value, namely, the interest heuristic model is continuously trained, then when training is completed, the optimal solution of the loss function is obtained, namely, the parameters of the interest heuristic model are trained to the optimal parameters, and the interest model is trained, so that the interest model can be accurately executed according to the user characteristic of the user heuristic interest information.

Referring to fig. 3, an embodiment of the present application provides a data processing method capable of determining an unknown interest of a user so as to push content for the user according to the unknown interest of the user. As shown in fig. 3, in particular, the execution subject of the method may be the server described above, and the method includes: s301 to S305.

S301: and obtaining a first score determined by the interest heuristic model to be trained according to the user characteristic data.

S302: and acquiring the intimacy degree between the user and each friend and the interest information of each friend.

The server records the friend relation of the user, wherein the friend relation of the user comprises user identifications of each user belonging to the friend relation with the user. The intimacy degree between the user and the friend can reflect the interaction frequency between the user and the friend and the intimacy of the friend relationship, which is equivalent to classification of the social relationship of the user, specifically, the intimacy degree can be a parameter value, and the larger the parameter value is, the higher the intimacy degree between the user and the friend is, and the higher the interactivity is.

As one implementation, the degree of intimacy between a user and various friends may be determined by the user's social information. The social information of the user comprises interaction information of the user and each friend, wherein the interaction information comprises forwarding times, collection times, comment times and grouping information, specifically, the forwarding times, collection times and comment times can be times of forwarding, collecting and comment of the content posted by the friend by the user, and the grouping information can be keywords of a plurality of groups established by the user and user identifiers in the groups.

As an implementation manner, parameters may be set for the forwarding times, the collection times, the comment times and the grouping information, that is, the parameters include a first parameter, a second parameter, a third parameter and a fourth parameter, where the first parameter corresponds to the forwarding times, the second parameter corresponds to the collection times, the third parameter corresponds to the comment times, and the fourth parameter corresponds to the grouping information. For ease of calculation, the first parameter, the second parameter, the third parameter, and the fourth parameter may all be normalized to the [0,1] value range interval.

Specifically, the forwarding times of the user to each user are obtained, namely the forwarding times corresponding to each friend are determined, then the forwarding times of all friends are added to obtain the total forwarding times, and then the forwarding times of each friend are divided by the total forwarding times to obtain a numerical value which is a first parameter of each friend. Similarly, the second parameter and the third parameter may be obtained.

The fourth parameter may be obtained by determining a grouping of the user on each friend, determining a keyword of each grouping, determining a category corresponding to the keyword of the grouping, and determining a first value corresponding to the keyword of the grouping according to scores corresponding to different preset categories, where the scores corresponding to the different categories are different, and some categories, such as a friend, have a higher set score, and for friends not grouped, the score of the default grouping is lower, so that the first value can represent whether the grouping in which the friend is located is a grouping corresponding to the friend with higher intimacy of the user. The first value is normalized and then used as a fourth parameter.

Then, the intimacy degree between the user and each friend is obtained according to the first parameter, the second parameter, the third parameter and the fourth parameter, and as an implementation mode, the first parameter, the second parameter, the third parameter and the fourth parameter can be summed, and the summed result is used as the intimacy degree.

As an embodiment, the social information of the user may be used as input of the graphSage model, and the friends of the user may be classified by the graphSage model, for example, the degree of intimacy between friends at the same job level may be close, the proportion of the identity may be close, and the proportion of the interest may be close.

In addition, the interest information of the friends can be counted in advance by the server, namely, each friend corresponds to one interest information, and the interest information can be further known interest information.

S303: and acquiring the weight value determined by the joint probability model according to the intimacy degree of each friend, the interest information of each friend and the sample data.

Specifically, if the sample data matches the interest information of most friends in friends with higher intimacy degree of the user, the determined weight value is smaller. That is, if most friends of the user with higher intimacy degree consider that a certain sample is of accurate interest, the output weight value is smaller.

According to the method, according to the intimacy degree of each friend, friends with intimacy degree larger than a certain threshold value with the user are searched for as candidate friends from a plurality of friends of the user, and then the number of friends matched with the sample is determined according to the interest information of each friend in the candidate friends, wherein the friends matched with the sample are the known interests of the friends with the sample being larger than a specified value based on the interest information of the friends. Then, a weight value is determined according to the number of friends matched with the sample, for example, the number of friends matched with the sample is recorded as a first number, the number of candidate friends is taken as a second number, and the ratio of the first number and the second number is taken as the weight value.

As another embodiment, the joint probability model is used to: the interest degree of each friend on the sample data is determined according to the known interest information of each friend; and determining the weight value according to the intimacy degree and the interestingness degree of the friends, wherein the greater the intimacy degree is, the greater the interestingness degree is, and the smaller the determined weight value is.

Specifically, the interest degree of each friend on the sample data is determined based on the known interest information of each friend, wherein the interest degree may be the possibility that the sample belongs to the known interest information of the friend, then, the intimacy degree between the friend and the user is determined, a weight value is determined according to the possibility and the intimacy degree, for example, a product of the acquired possibility and the intimacy degree is recorded as a reference result, then, a reference result corresponding to each friend is acquired, then, a weight value is obtained according to the reference result of each friend, for example, an average value of all the reference results is acquired, and the average value is taken as the weight value.

As an embodiment, the joint probability model may also be trained based on known interest information of the user, in particular, please refer to the following examples.

As an implementation, the structure of the joint probability model is shown in FIG. 4, where the friend relationship data may be the social information, and the friend partition model may be a GraphSage model, which is used for outputting User unbinding, where the User unbinding is equivalent to indicating the intimacy between the User and other users, that is, the intimacy of friends. And obtaining the weight value through three-layer full-connection calculation. As an embodiment, the three fully connected layers may be multi-layer perceptrons (Multilayer Perceptron, MLP), specifically, the first fully connected layer is an input layer of the MLP, the second fully connected layer is a hidden layer of the MLP, and the third fully connected layer is an output layer of the MLP. Wherein, the loss function of the hidden layer may be:

wherein x is User subeding.

S304: and adjusting the first score according to the weight value, wherein the weight value is positively correlated with the first score.

S305: and training the interest heuristic model to be trained according to the adjusted first score.

Referring to fig. 5, an embodiment of the present application provides a data processing method capable of determining an unknown interest of a user so as to push content for the user according to the unknown interest of the user. As shown in fig. 5, in particular, the execution subject of the method may be the server described above, and the method includes: s501 to S507.

S501: and obtaining a first score determined by the interest heuristic model to be trained according to the user characteristic data.

Referring to FIG. 6, the principles of the interest heuristic model are described in connection with the interest heuristic model structure shown in FIG. 6.

As shown in FIG. 6, the user characteristic data entered into the interest heuristic model includes attribute characteristics, base characteristics, short-term behavioral characteristics and behavioral characteristics. The attribute features of the user include basic features such as gender, age, region and app of the user, the basic features of the user include clicking interesting behavior features of the user for content products displayed on the specified interface, specifically, the clicking interesting behavior features may be feature data of content clicked by the user on the specified interface at high frequency, where the feature data of the content may include data such as description information and category of the content, and the description information may include information such as keywords of the content. As an embodiment, the specified client may be a WeChat APP, the specified interface may be a WeChat glance interface, and the plurality of contents may be contents that can be displayed in the glance interface. That is, the basic feature is data generated by the user for the operation of displaying the content at the specified interface of the client.

The short-term behavior feature and the behavior feature may be data generated by the operation of the user on the display content of the other interface belonging to the client together with the specified interface, that is, the short-term behavior feature and the behavior feature are data generated by the operation of the display content displayed by the user on the other interface of the client. As one embodiment, both the short-term behavioral characteristics and the behavioral characteristics are generated based on the user's manipulation of the displayed content on the other interface, and the behavioral characteristics characterize a long-term point of interest, e.g., the user's content of interest or praise on the other interface, while the short-term behavioral characteristics characterize the user's specific operational behavior on the other interface. For example, the user clicks content 1, content 2, content 3, content 1 and content 2 in sequence on other interfaces within a certain period of time, then the behavior feature records that the user clicks content 1 twice, content 2 twice and content 3 once in the period of time, and the short-term behavior feature records the whole clicking operation of the user, namely, records not only the number of clicks but also the sequence of clicks. As an embodiment, the interface is designated as a glance interface of the WeChat, while the other interfaces may be interfaces of applets, interfaces of public numbers, etc., and the short-term behavioral characteristics and behavioral characteristics may be behavioral characteristics of the user on the WeChat (e.g., reading articles, glancing the number of lived, open public numbers, applet usage, short-term reading history, etc.).

As shown in fig. 6, by inputting respective feature vectors, each feature vector corresponds to one domain, for example, the user feature data includes an attribute feature, a basic feature, a short-term behavior feature, and a behavior feature, and the corresponding domain is four domains. Discrete features s= (S1, S2 …, sn) of each domain are transformed into a continuous vector W by embedding _e ∈R ^m*D Where m is the dimension of the embedded vector, D is the size of the dictionary, and the domains are not shared with each other. Vector W after each domain is embedded _e By means of a mean function, i.e. average_pooling (W _e ) An operation of converting the characteristics of each domain into a vector expression, namely E, through a mean value pooling operation _e ＝Average_Pooling(W _e ). Thereafter, feature fusion, i.e., C, is performed by a feature stitching function (i.e., concat) operation _e12 ＝Concat(E _e1 ，E _e2 ). Then, the nonlinear characteristic expression change is captured through a plurality of full-connection layers, namely, the nonlinear characteristic expression change is obtained through a linear rectification function, namely, R _e ＝Relu(E _e ). Dot product is then calculated with interest tags, i.e. Dot _e1,e2 ＝Dot(E _e1 ,E _e2 ). Specifically, the process of calculating the dot product may be understood as calculating the similarity between the interest tag shown in fig. 6 and the user feature data, and determining the estimated interest information of the user according to the similarity. For example, if the similarity between the user feature data and the cartoon label is higher, the user's interest in the cartoon can be indicated to be higher, and the cartoon can be used as estimated interest information of the user.

In addition, the Attention mechanism shown in fig. 6 is an Attention Model (Attention Model), and is essentially derived from the human visual Attention mechanism. People's vision generally does not see a scene from the head to the tail all at a time while perceiving something, but often looks at a specific part of the attention as required. And when people find that a scene often appears something that the people want to observe on a certain part, people can learn to pay attention to the part when similar scenes appear again in the future, focus more attention on a useful part, and the essence of an attention model is weighting, namely an allocation mechanism of weight parameters, and the aim is to assist the model to capture important information. The Attention model may calculate a weight for each input feature and then weight the feature, where the greater the weight of the feature, the greater the contribution of the feature to the current recognition result. For example, given a set of < keys, value > and a target (query) vector query, the similarity between the query and each set of keys is calculated to obtain the weight coefficient of each key, and the value is weighted and summed to obtain the final output result.

Therefore, the interest heuristic model can obtain estimated interest information of the user according to the user characteristic data. It should be noted that, when the interest heuristic model is not trained, that is, the accuracy of the estimated interest information output by the initial interest heuristic model is poor, the accuracy of the estimated interest information output by the interest heuristic model needs to be improved in a training manner. In order to improve the accuracy of the estimated interest information output by the interest heuristic model and improve the accurate dividing line of the known interest and the unknown interest of the user, the embodiment of the application can train the interest heuristic model by adopting the combined training model shown in fig. 7.

S502: and acquiring a weight value determined by the joint probability model according to the sample data.

S503: and adjusting the first score according to the weight value.

S504: and training the interest heuristic model to be trained according to the adjusted first score.

S505: and acquiring a second score determined by the accurate interest model according to the known interest information of the user.

Wherein the second score is used to characterize a likelihood that the sample data belongs to a known interest of the user, i.e. a second likelihood, the known interest information being a determined interest preference of the user.

As an embodiment, the known interest information may be an interest determined according to click data of the pushed history content by the user. Specifically, the known interest information of the user is used to characterize the interest tags of the user that the server has currently determined. For example, the server stores a user identifier of each user and known interest information corresponding to the user identifier. As one embodiment, the known interest information may be an interest tag, and when the server recommends a content for the user according to the known interest information of the user, the interest tag of the user may be determined, and a content matching the interest tag is searched for from a plurality of contents as a recommended content.

In some embodiments, the known interest information is an interest tag determined from user click data on pushed historical content. In some embodiments, the server counts click data of the user on each content for a preset period of time. Wherein the click data includes a click operation. Then, determining the content clicked by the user in a preset time period, analyzing the clicked content, extracting key features of the content, and counting and partially ordering each key feature, so that the known interest information of the user is obtained. For example, the number of clicks of the durian-related content by the user is relatively high within a preset period of time, so that durian can be determined as an interest tag of the user, that is, the known interest information of the user includes durian.

As another implementation, the known interest information may also be information entered by the user. For example, the user sends interest information to the server, which takes the interest information sent by the user as known interest information and binds with the user identification of the user. The method for inputting the interest information by the user can be that the server provides an interest submitting interface for the user, an interest inputting area exists in the interface, the user can input the interest information in the area through a virtual keyboard, and the user can select the interest from a plurality of interest labels in the area to serve as the interest information input by the user.

In some embodiments, when a user registers on a client, in the process of registering on the client, the client displays an interest submitting interface, and the user inputs interest information in the interest submitting interface to complete registration, so that the client sends the interest information input by the user and a user account registered by the user to a server, and the server takes the interest information submitted by the user as known interest information and stores the user account in correspondence with the known interest information of the user.

The estimated interest information is known interest information not belonging to the user, but the likelihood of the user being interested is relatively high. For example, if the estimated interest information and the known interest information are both interest tags, the estimated interest information is a tag that is not a known interest tag of the user, but the likelihood of interest of the user is relatively high. For example, the known interest tags of the user may include durian, the estimated interest information of the user may include pineapple, the estimated interest information of the user may be used to characterize information that may be of interest to the user, or the interest tags may have a degree of interest greater than a specified threshold.

As an embodiment, the known interest information of the user may be obtained based on a precise interest model, and specifically, the precise interest model may be a FNN, RNN or other model, where the precise interest model obtains the known interest information of the user according to a user portrait, and the user portrait may include user basic features, behavior features, content, contextual features, and the like.

Similar to the interest heuristic model, the accurate interest model may also determine a likelihood that the sample data belongs to the user's known interest, i.e., a second likelihood, based on the user's known interest information.

S506: and adjusting the weight value according to the first score and the second score.

S507: and training the joint probability model according to the adjusted weight value.

As one implementation, the joint training model uses the accurate interest model, the interest heuristic model and the joint probability model to train the interest heuristic model and the joint probability model respectively, so that when the interest heuristic model is trained by using the whole joint training model, the accuracy of the interest information output by the interest heuristic model influences the convergence speed of the whole model, and further influences the training speed. Therefore, the method ensures that the performance of the interest heuristic model is not too weak when the joint model is trained, so that the learning fluctuation and the convergence speed of the whole model are slow, and the accuracy of the interest heuristic model is improved, so that a more proper joint model training direction can be provided, a stable and effective model can be obtained, the convergence speed of the model is improved, and the interest heuristic model can be trained before the joint model is trained.

In the embodiment of the present application, the interest heuristic model that is not started to be trained is named as initial interest heuristic model training, as shown in fig. 6, after the initial interest heuristic model is constructed, user feature data is input into the initial interest heuristic model, and then the initial interest heuristic model outputs initial interest information, where the initial interest information characterizes the output result of the initial interest heuristic model that is not trained.

As one embodiment, sample data is obtained and the sample data and the output of the initial heuristic model of interest (i.e., the output of the dot product function described above) are trained by a loss function. Wherein the loss function is used for model fitting, and the degree of fitting is usually measured by the loss function. Minimizing the loss function means that the fitting degree is best, and the corresponding model parameter is the optimal parameter. As an embodiment, the initial interest information described above can be optimized by optimizing the parameters of the interest heuristic model using a gradient descent algorithm on the loss function.

As an embodiment, the sample data may be all content pushed to the user through the designated interface for a preset period of time. As another embodiment, the sample data may be content clicked by the user within the specified interface for a preset period of time.

As one implementation, a sigmoid loss is calculated, and the initial interest heuristic model is trained 5000 times specifically by back propagation optimization loss, so as to obtain the interest heuristic model to be trained. After pre-training, the interest heuristic model learns and trains the heuristic interests of the user, and is used as a representation for describing the strength of the heuristic interests of the user, so as to prepare for the training of the combined model. It should be noted that, the interest heuristic model to be trained is training of the initial interest heuristic model, and the interest information output by the interest heuristic model to be trained can be named as interest information to be confirmed, that is, the accuracy of the interest information to be confirmed is still not ideal, and the interest heuristic model to be trained needs to be trained through the joint training model.

As another embodiment, the initial heuristic model of interest may not be pre-trained. Specifically, after the initial interest heuristic model is produced, the initial interest heuristic model is directly trained by the joint training model, and then the interest heuristic model to be trained shown in fig. 7 is an initial interest heuristic model which is not trained in advance, rather than an initial interest heuristic model which is trained in advance.

As shown in fig. 7, fig. 7 shows a structure of a joint training model, the joint model training is introduced into a precise interest model of a recommendation system, the joint probability model is used for joint training of the precise interest model and an interest heuristic model, and unified training is performed on sample data of a user. The accurate interest model is responsible for solving the known interest depiction of the user in the sample, the interest heuristic model is responsible for solving the unknown interest calculation of the user, and the accurate calculation log of the joint probability model and the division of the heuristic sample are accurate.

As shown in FIG. 7, the accurate interest model outputs known interest information corresponding to the user. As one implementation mode, the accurate interest model is a model which is already trained, so that the accurate interest model does not need to be trained through a combined training model, and because the accurate interest model is introduced, the interest heuristic model to be trained can be trained according to the sample data and the known interest information corresponding to the user, so that the interest information to be confirmed is optimized to be estimated interest information of the user.

Since the known interest information of the user output by the accurate interest model reflects the interest of the user which is already determined, namely, the content which is similar to the known interest information is the content which is interested by the user, the content which is similar to the content can also be the content which is interested by the user, for example, the known interest information of the user comprises shooting games, and the user can also like moba games, and the correlation between the two can be used as the basis for measuring whether the moba games are the unknown interests of the user, namely, the estimated interest information. Therefore, the interest information to be confirmed can be optimized by referring to the known interest information corresponding to the user and training the interest heuristic model to be trained according to the sample data.

As an embodiment, as shown in fig. 7, the user data 1 for inputting the interest heuristic model may include search terms, operation data of the user on the applet, operation data of the user on the client, and sequence behavior, the sequence behavior may include clicking behavior of the user on a glance or other interface, specifically, the user data may refer to the user feature data for inputting the interest heuristic model, and the user data 2 for inputting the precise interest model may include reading article data corresponding to the user, clicking video data, focused public numbers, and other reading histories.

The sample data is input into the accurate interest model, the interest heuristic model and the joint probability model, the accurate interest model can determine a second score of the sample data belonging to the known interest of the user according to the known interest information corresponding to the user, the interest heuristic model can determine a first score of the sample data belonging to the unknown interest of the user, and the joint probability model outputs a weight value of the sample. And then training the interest heuristic model to be trained according to the first score, the second score and the weight value, wherein the weight value can correct the first score determined by the interest heuristic model to be trained so as to optimize the accuracy of the first score.

As an embodiment, the loss transfer value of the interest heuristic model is used to represent the difference between the sample of unknown interest to the user estimated by the interest heuristic model and the sample of unknown interest to the real user, i.e. the difference can represent the difference between the predicted value, which is the first likelihood of the interest heuristic model being output, i.e. the first score, and the real value, which represents the real likelihood that the sample is of unknown interest to the user.

In order to obtain the unknown interest information of the user by combining the known interest information of the user, determining a loss transfer value of the interest heuristic model according to the first probability, the second probability and the weight value. Specifically, the first likelihood is a first score, the second likelihood is a second score, and the embodiment of determining the loss transfer value of the interest heuristic model according to the first likelihood, the second likelihood and the weight value may be that a total score is obtained according to a weight value, the first score and the second score, the weight value being a likelihood that the sample belongs to a known interest of a user according to a joint probability model; and determining a loss transfer value of the interest heuristic model according to the total score.

Specifically, the embodiment of determining the total score according to the weight value, the first score, and the second score may be determining the total score according to the following formula (1).

score＝(1-d _s )*score _p +d _s *score _ep (1)

Wherein score is the total score, d _s Score as a weight value _p Score of the second score _ep Is the first score. As one embodiment, score, d _s 、score _p And score _ep The value ranges of the (E) are all 0,1]。

In the embodiment of the application, the weight value is a probability that the sample belongs to the known interest of the user, which is obtained according to the joint probability model, that is, the weight value provides a correction force, so as to correct the score of the interest heuristic model to the sample.

As one implementation, the joint probability model can determine the probability that the currently trained sample belongs to the precise interest model. It should be noted that, the second score output by the accurate interest model is used to characterize the possibility that the sample data belongs to the known interest of the user, and the higher the score is, the more likely the sample is to match the known interest of the user, and the weight value output by the joint probability model characterizes the probability that the sample belongs to the accurate interest model, and the meaning expressed by the weight value is different from the second score, because the sample belongs to the accurate interest model and does not represent that the sample is necessarily the known interest of the user, and the influence of the score of the accurate interest model on the sample on the finally determined unknown interest of the user is expressed by the weight value.

The training process of the joint training model will be described below with reference to fig. 7.

Specifically, bp shown in fig. 7 is back propagation (Backpropagation), fw is an abbreviation of forward, and represents model prediction. The joint training model is divided into two parts, namely interest heuristic model training and joint probability model training. Each model inputs corresponding user interest vector, and after feature space calculation conversion of each model, the score of each model to the current sample is output, namely the second score _p First score _ep And weight value d _s 。

As shown in fig. 7, the data input to the joint probability model is a user vector, which may be a high-dimensional image representation of the user, and the spatial vector may be a spatial transformation of the user vector as an input to the joint probability model. Specifically, according to the type of content to be pushed or specific pushing requirements, the user vector may be set according to the specific pushing requirements, and in the embodiment of the present application, the user vector may be social information of the user.

And then calculating according to the formula (1) to obtain the total score, the back propagation optimization joint probability model and the interest heuristic model parameters.

For the interest heuristic model, its loss function loss is:

Wherein θ _j For the parameters of the interest heuristic model, the formula (2) is a gradient descent algorithm, and the interest heuristic model is trained on the scores of the samples through the gradient descent algorithm to obtain an optimal solution, so that the trained interest heuristic modelThe model can accurately output estimated interest information of the user. It can be seen that when d _s When smaller, more belongs to the accurate model, that is, the probability that the sample is suitable for training the interest heuristic model is smaller, loss transmission of loss to the interest heuristic model is smaller, and it can be seen that the weight value makes the overall score of the right formula smaller, that is, the weight value makes the first score output by the interest heuristic model smaller. When d _s When the probability that the sample is suitable for training the interest heuristic model is larger, loss transmission of loss to the interest heuristic model is larger, the weight value enables the overall score of the right formula to be larger, namely the weight value enables the first score output by the interest heuristic model to be larger.

For the joint probability model, the loss function loss is:

wherein θ _i For the parameters of the interest heuristic model, the formula (3) is a gradient descent algorithm, and the sample is scored and trained by the joint probability model through the gradient descent algorithm to obtain an optimal solution, so that the trained joint probability model can accurately determine the weight value, namely the possibility that the current sample belongs to the accurate interest model can be accurately determined. When score _p And score _ep When the difference is small, the boundary between the accurate interest model and the interest heuristic model is blurred, loss transfer of loss to the joint probability model is small, that is, the weight value is used to characterize the probability that the sample is suitable for training the interest heuristic model, if one sample gives a higher score to both the accurate interest model and the interest heuristic model, the weight value should be small when the difference between the accurate interest model and the interest heuristic model is small, because the sample may be more suitable for training the accurate interest model, so score _p And score _ep The difference results in a smaller weight value, thereby reducing the likelihood that the sample is suitable for training the interest heuristic model, that is, when the boundaries of the accurate interest model and the interest heuristic model are blurred,the weight value is reduced when it is difficult to accurately divide the sample for training of a precise interest model or an interest heuristic model. Similarly, when score _p And score _ep When the difference value is large, the boundaries of the accurate interest model and the interest heuristic model are clear, loss transmission of loss of the loss to the joint model is large, and the probability that the sample is suitable for training the interest heuristic model is large. Therefore, through the formula (3), the joint probability model can be trained by using the accurate interest model and the interest heuristic model, so that the weight value determined by the joint probability model is more accurate. In particular, the parameters in the output f (w1x+b1) of the hidden layer in the joint probability model may be trained such that the parameters get an optimal solution, e.g. a gradient descent algorithm may be employed.

Referring to fig. 8, an embodiment of the present application provides a data processing method capable of determining an unknown interest of a user so as to push content for the user according to the unknown interest of the user. As shown in fig. 8, in particular, the execution subject of the method may be the server described above, and the method includes: s801 to S807.

S801: and obtaining a first score determined by the interest heuristic model to be trained according to the user characteristic data.

S802: and acquiring a weight value determined by the joint probability model according to the sample data.

S803: and adjusting the first score according to the weight value.

S804: and training the interest heuristic model to be trained according to the adjusted first score.

S805: obtaining estimated interest information of the user according to the trained interest heuristic model.

The interest heuristic model is obtained according to the above data processing method, and specifically please refer to the above embodiment, which is not described herein.

S806: and determining the content to be pushed according to the estimated interest information of the user.

As one implementation, the content to be pushed may be determined from a plurality of content based on the estimated interest information of the user.

An information tag for each content is determined, the information tag describing the content. In some embodiments, the information tag may be a category of content, a keyword of content, and the like.

Wherein the plurality of contents are contents for display through a designated interface of the designated client. In one embodiment, the specified client may be a client for displaying the content to be pushed by a user terminal, and the server may be a server corresponding to the specified client, and the user identifier of the user may be a user account logged in to the specified client. As shown in fig. 9, the designated interface is an information presentation interface of the designated client, in which a plurality of contents, such as a content 301, a content 302, a content 303, and a content 304 shown in fig. 9, can be displayed, and the user can review the content 301, the content 302, the content 303, and the content 304 in the designated interface, and can click on each of the contents, so as to be able to enter a detail interface corresponding to each of the contents, such as a detail interface corresponding to the content 302 shown in fig. 10, and when the user clicks on the content 302 in the designated interface shown in fig. 9, the detail interface of the content shown in fig. 9 is displayed on the screen, and the user review the detail content corresponding to the content 302 in the interface shown in fig. 10.

And determining the implementation mode of the content to be pushed from the plurality of contents according to the estimated interest information of the user, acquiring an information label of each content, matching the information label with the estimated interest information of the user, and taking the content corresponding to the information label matched with the estimated interest information as the content to be pushed. The matching method may be to obtain the similarity between the information tag and the estimated interest information of the user, and use the content with the similarity greater than the specified threshold as the content to be pushed.

As an implementation manner, the matching degree of the information tag of each content and the estimated interest information of the user may be determined, the matching degree is ordered to obtain a push sequence, and N contents ordered in the push sequence to be the content to be pushed are used as the content to be pushed. Wherein N is a positive integer greater than 1.

It should be noted that, the content matching the known interest information of the user may be named as user accurate interest content, the content matching the estimated interest information of the user may be named as user trial interest content, the user accurate interest content is usually related to the content that the user has clicked or the content that has frequently clicked, and the trial interest content may be the content of the category that the user has not clicked or the content of the category that has not frequently clicked, but the content that the user estimated by the server may be interested or the content that the user may have a high probability of clicking.

S807: and pushing the content to be pushed to the client corresponding to the user.

The client corresponding to the user may be the above specified client.

As an implementation manner, the content to be pushed is also displayed in the above-mentioned designated interface of the designated client, and as an implementation manner, the designated interface is an interface of the designated application program in the client, and in the embodiment of the present application, the designated interface may be a "looking at" interface of the WeChat.

Specifically, the server pushes the content to be pushed to the appointed client of the user, and the appointed client displays the content to be pushed in an appointed interface.

In some embodiments, the specified client initiates the specified interface and displays the content to be pushed along with other content on the specified interface, where the other content may be push content determined by the server according to other policies, where the other policies may include push content determined according to known interest information of the user.

In other embodiments, when the server determines that the specified interface of the specified client has been opened, the specified interface may be pushed to the specified client and displayed within the specified interface if the specified interface has been opened by the user based on some push policy.

In one embodiment, the method for displaying the content to be pushed in the designated interface may be that after the designated client obtains the content to be pushed, the designated client waits for a next refresh operation of the user, and when the next refresh arrives, the content to be pushed is displayed in a designated area of the designated interface. In some embodiments, the designated area is a top area of a content display page within the designated interface, and the refresh operation is a user-triggered downward movement of a top edge of the page. Therefore, after the specified client acquires the content to be pushed, when the top edge of the user-triggered page moves downward and is maintained for a certain period of time, a page refresh operation is performed, and as shown in fig. 11, the content to be pushed is displayed in the top area of the page of the specified interface. Therefore, when the user refreshes, the content to be pushed can be immediately displayed to the user in the appointed interface, and the content to be pushed can be obtained in advance, so that the user terminal can generate the content to be displayed in advance according to the content to be pushed, and when the refresh request is obtained, the content to be displayed can be immediately displayed, and the waiting time required by refreshing the appointed interface is reduced.

In another embodiment, the embodiment that the content to be pushed is displayed in the designated interface may further be that after the designated client obtains the content to be pushed, the designated client waits for the user to open the designated interface next time, and then displays the content to be pushed in the designated interface.

As yet another embodiment, the content 301, the content 302, the content 303, and the content 304 on the designated interface shown in fig. 9 may be displayed by the server in the designated interface based on a request of a user or other pushing policy, and in the case that the designated interface is displayed, the server determines the content to be pushed by the user and sends the content to be pushed to the designated client. When the appointed client determines that the user clicks the content on the current appointed interface, the appointed client enters the detail interface of the selected content, and when the user returns to the appointed interface, the content to be pushed is displayed below the content selected by the user.

For example, the user clicks the content 302 in the interface shown in fig. 9, and then the designated client displays the detail interface of the content 302 shown in fig. 10, and when the user returns to the designated interface after closing the detail interface, the content displayed by the designated interface becomes the content shown in fig. 12, that is, the content to be pushed is displayed below the content 302 shown in fig. 9 and 12, so that the content to be pushed can be displayed below the content determined to be currently browsed by the user, on the one hand, the content to be pushed can be quickly and effectively pushed to the user, on the other hand, the situation that the content displayed by the designated interface in fig. 9 may be related to the known interest information of the user can be avoided, and the variability of the respective contents in fig. 9 may be small, and if the user continuously reads the content with the smaller variability, aesthetic fatigue may be caused.

For example, if the known interest information of the user includes a cartoon, the content displayed in fig. 9 is the content determined by the server according to the known interest information of the user, and then the user is in the designated interface in fig. 9, the content shown in fig. 9, even the content not shown in fig. 9, is related to the cartoon, and the user continuously sees that a plurality of contents are related to the cartoon. Therefore, after the content to be pushed is determined according to the estimated interest information of the user, the content to be pushed is displayed below the content selected by the user of the designated interface, so that the user can touch the content which is beyond some cartoon and has high possibility of being interested, and the user experience is improved.

In one embodiment, after determining the content to be pushed, the server may preset a pushing condition, and when the parameter information of the server meets the pushing condition, the server performs pushing the content to be pushed to the specified client of the user, so that it can be ensured that the server can push the content to be pushed to the specified client corresponding to the target user only when the parameter information meets the pushing condition.

The push condition is a time interval, and the parameter information may be a system time corresponding to the server, that is, a current time. The server acquires the current system time as the current time, then the time of last sending the activity information to the user is used as the historical time, the time difference between the historical time and the current time is acquired, whether the time difference is larger than or equal to a specified time interval threshold value is judged, and if the time difference is larger than or equal to the specified time interval threshold value, the operation of acquiring the activity information to be pushed is executed. The specified time interval threshold may be set in advance according to a user requirement or a push requirement, for example, may be 24 hours, so that it can be ensured that the active content is pushed once a day.

In addition, the pushing condition is not only the time interval, but also the network parameter between the server and the user terminal meets the specified communication condition, so that the problem that when the network state between the server and the user terminal is poor, the active information to be pushed is still sent to the user terminal, and the resource waste is caused can be avoided. Specifically, the server acquires communication parameters between the server and the user terminal corresponding to the user to be pushed, determines whether the communication parameters meet specified communication conditions, and pushes the activity information to be pushed to the user terminal corresponding to the user to be pushed if the specified communication conditions are met.

The communication parameter may specifically be a channel quality, wherein the channel quality may be an error vector magnitude of a channel, the number of access points, a signal strength, etc. The error vector magnitude (Error Vector Magnitude, abbreviated as EVM) refers to a vector difference between an ideal error-free reference signal and an actual transmission signal at a given moment, and is used for measuring an amplitude error and a phase error of a modulation signal, and the EVM specifically refers to a degree of proximity between an IQ component generated when a receiving terminal demodulates the signal and an ideal signal component, which is an index for considering the quality of the modulation signal. The smaller the EVM, the better the channel quality of the channel. The number of the access points can be obtained when the channels are scanned, so that the number of the access points on each channel can be determined, and the more the number of the access points is, the worse the channel quality is, and the better the reverse is. Similarly, the signal strength can be obtained when the channel scans, and the higher the signal strength is, the higher the channel quality is, and the lower the channel quality is otherwise.

The specific embodiment of determining whether the communication parameter satisfies the specified communication condition is to determine whether the channel quality satisfies the specified communication quality, and if so, determine that the communication parameter satisfies the specified communication condition. Specifically, the channel quality is the error vector magnitude of the channel, and if the error vector magnitude of the channel is smaller than a specified value, it is determined that the channel quality satisfies the specified communication quality, and further it is determined that the communication parameter satisfies the specified communication condition.

Thus, the unknown interests of the user are determined from the known interest information. Then, determining content to be pushed according to the estimated interest information of the user; the content to be pushed is pushed to the client corresponding to the user, so that the content pushing method and the content pushing device not only can push the content for the user according to the known interests of the user, but also can push the content for the user according to the unknown interests of the user, and the diversity of the pushed content is improved.

Referring to fig. 13, an embodiment of the present application provides a data processing method capable of determining an unknown interest of a user so as to push content for the user according to the unknown interest of the user. As shown in fig. 13, specifically, the execution subject of the method may be the server described above, and the method includes: s1301 to S1308.

S1301: and obtaining a first score determined by the interest heuristic model to be trained according to the user characteristic data.

S1302: and acquiring a weight value determined by the joint probability model according to the sample data.

S1303: and adjusting the first score according to the weight value.

S1304: and training the interest heuristic model to be trained according to the adjusted first score.

S1305: obtaining estimated interest information of the user according to the trained interest heuristic model.

S1306: content meeting the specified requirements is determined as alternative content.

As one embodiment, the server is provided with a content database, in which a plurality of contents are stored, and the plurality of contents are displayed for being sent to clients corresponding to respective users. The method provided by the embodiment of the application needs to push the content for the user by a plurality of contents in the content database.

Specifically, the plurality of contents are recorded as candidate contents, and from the candidate contents, contents meeting specified requirements are determined as candidate contents.

In one embodiment, the estimated interest information of the user may be a plurality of estimated interest tags, and considering that the number of estimated interest tags of the user is relatively large, if the content to be pushed is determined based on the too many estimated interest tags, the types or tags of the content seen by the user are too scattered, so that the possibility that the user clicks on a certain content is not high, and the accuracy of pushing is reduced. A screening condition may be set by which the alternative content is determined.

As an embodiment, the screening condition may be determined according to the current geographical location of the user terminal. Specifically, the server acquires current position information of the user terminal, determines a position area in which the current position information is seated, then determines shops in the position area, and determines shop information corresponding to each shop, wherein the shop information can be description information of commodities sold by the shops, the description information can include types of the commodities, and then determines content matched with the commodity information from the to-be-selected content as an alternative content. Specifically, the shop information of each shop in the location area where the user terminal is located is matched with each content in the content to be selected, and all the matched contents are used as candidate contents. For example, the shop information of shops near the current geographic position of the user terminal comprises fruits, toys, movie theatres, restaurants and the like, then the server takes the content related to the shop information of the fruits, toys, movie theatres, restaurants and the like in the content to be selected as the candidate content, and then screens the subsequent content to be pushed from the candidate content.

As another embodiment, the embodiment of determining the content meeting the specified requirement as the alternative content may also be that the content matching the known interest information of the user is determined from among the plurality of contents; and taking the content except the matching content as the candidate content in the plurality of contents.

Specifically, as shown in fig. 14, known interest information of the user is acquired, and content matching the known interest information of the user is determined from the above-mentioned candidate content according to the known interest information of the user as accurate interest content. Then, the content other than the precise interest content in the candidate content is taken as the candidate content. Therefore, the estimated interest information determined according to the interest heuristic model can be prevented from being too similar to the known interest information of the user, so that the pushing content determined based on the estimated interest information is too repeated with the content determined based on the known interest information, and the excessive repeated content is pushed to the user. Therefore, when pushing content is determined based on the estimated interest information, the content belonging to the accurate interest is directly filtered, so that the situation that the content is not pushed accurately due to unreasonable weight value setting can be avoided.

S1307: and determining the content to be pushed from the alternative content according to the estimated interest information of the user.

S1308: and pushing the content to be pushed to the client corresponding to the user.

In addition, as shown in fig. 14, after pushing the content to be pushed to the client corresponding to the user, operation data of the user on the content to be pushed may also be collected, and the operation data continues to be used as user feature data, and then the interest heuristic model is trained according to the method mentioned in the above embodiment, so as to further optimize the estimated interest information output by the interest heuristic model.

As an implementation manner, the embodiment of the application can also push the content determined based on the known interest information of the user, namely the accurate interest content, to the client corresponding to the user. Specifically, the content to be pushed, which is determined according to the estimated interest information of the user, is named as unknown interest content, and the accurate interest content and the unknown interest content can be simultaneously pushed to the client corresponding to the user according to a predetermined strategy.

As an embodiment, the content 301, the content 302, the content 303, and the content 304 shown in fig. 9, 11, and 12 are all precise interest contents, and as shown in fig. 11, the unknown interest contents may be displayed when the user refreshes the designated interface, for example, the unknown interest contents are displayed in the top area of the page of the designated interface, or as shown in fig. 12, when the user clicks a certain precise interest content, the unknown interest contents are displayed in the designated interface, and specifically, the unknown interest contents are displayed in the adjacent area below the clicked precise interest content. As shown in fig. 11 and 12, both the content 501 and the content 502 are unknown interest content. Therefore, when a user opens a designated interface, the server pushes the accurate interest content to the client of the user and displays the accurate interest content in the designated interface, so that the reading interest of the user can be mobilized through the accurate interest content, and then after the user clicks a certain accurate interest content, the unknown interest content is pushed for the user, so that the user can click the unknown interest content more likely under the fumigating of the accurate interest content. As another embodiment, the precise interest content and the unknown interest content can be displayed in a staggered manner as shown in FIG. 15.

As one implementation, the number of precise and unknown content of interest may be set based on the total number of push content required. Specifically, assuming that the content pushed for the user at a time is not more than 20, the total number of required pushed contents is 20, and the ratio between the precise interest content and the unknown interest content is set to be M1/M2, so that the precise interest content is 20×m1/(m2+m1), and the number of the unknown interest content is the difference between the total number and the number of the precise interest content.

Therefore, by determining the content matching the known interest information of the user from among the plurality of contents, taking the content other than the matching content as the candidate content, and then determining the content to be pushed from among the candidate content according to the estimated interest information of the user, it is possible to avoid that the content pushed based on the estimated interest information of the user has too much repeated content with the pushed content determined based on the known interest information of the user.

Referring to fig. 16, an embodiment of the present application further provides a data processing apparatus, where the data processing apparatus 1600 includes: a first acquisition unit 1601, a second acquisition unit 1602, an adjustment unit 1603, and a training unit 1604.

A first obtaining unit 1601, configured to obtain a first score determined by the interest heuristic model to be trained according to the user feature data, where the first score is used to characterize a likelihood that the sample data belongs to an unknown interest of the user.

A second obtaining unit 1602, configured to obtain a weight value determined by the joint probability model according to the sample data, where the weight value is used to characterize a likelihood that the sample is suitable for training the interest heuristic model.

Further, the second obtaining unit 1602 is further configured to obtain intimacy degree between the user and each friend and interest information of each friend; and acquiring the weight value determined by the joint probability model according to the intimacy degree of each friend, the interest information of each friend and the sample data.

Further, the second obtaining unit 1602 is further configured to determine, according to the known interest information of each friend, an interest degree of each friend on the sample data; and determining the weight value according to the intimacy degree and the interestingness degree of the friends, wherein the greater the intimacy degree is, the greater the interestingness degree is, and the smaller the determined weight value is.

An adjusting unit 1603, configured to adjust the first score according to the weight value, where the weight value is positively correlated with the first score.

The training unit 1604 is configured to train the interest heuristic model to be trained according to the adjusted first score, where the trained interest heuristic model is configured to determine estimated interest information of a user according to the user feature data, where the estimated interest information is unknown interest information of the user, and the estimated interest information is configured to determine content to be pushed to a client corresponding to the user.

Further, the data processing apparatus 1600 further includes: the joint training unit is used for acquiring a second score determined by the accurate interest model according to the known interest information of the user, wherein the second score is used for representing the possibility that the sample data belong to the known interest of the user, and the known interest information is the determined interest of the user; adjusting the weight value according to the first score and the second score, wherein the weight value is used as an output value of a loss function of the joint probability model; and training the joint probability model according to the adjusted weight value.

Further, the data processing apparatus 1600 further includes: and the pre-training unit is used for pre-training the initial interest heuristic model for a designated number of times according to the sample data and the user characteristic data to obtain the interest heuristic model to be trained.

Specifically, the user characteristic data includes basic characteristics and behavior data of a user, wherein the behavior data is data generated by the operation of a specified application module of a client and other application modules, and the other application modules are application modules related to the specified application module and belong to the client.

Further, the data processing apparatus 1600 further includes: the pushing unit is used for obtaining estimated interest information of the user according to the trained interest heuristic model; determining content to be pushed according to the estimated interest information of the user; and pushing the content to be pushed to the client corresponding to the user.

Specifically, the pushing unit is further configured to determine content meeting the specified requirement, as an alternative content; and determining the content to be pushed from the alternative content according to the estimated interest information of the user.

Specifically, the pushing unit is further configured to determine, from among the plurality of contents, a content matching the known interest information of the user; and taking the content except the matching content as the candidate content in the plurality of contents.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the apparatus and modules described above may refer to the corresponding process in the foregoing method embodiment, which is not repeated herein.

Referring to fig. 17, a block diagram of an electronic device according to an embodiment of the present application is shown. The electronic device 10 may be a smart phone, tablet, electronic book, or other electronic device capable of running applications. Specifically, in the embodiment of the present application, the electronic device 10 may be the server 200 described above.

The electronic device 10 of the present application may include one or more of the following components: a processor 110, a memory 120, and one or more application programs, wherein the one or more application programs may be stored in the memory 120 and configured to be executed by the one or more processors 110, the one or more program(s) configured to perform the method as described in the foregoing method embodiments.

Processor 110 may include one or more processing cores. The processor 110 utilizes various interfaces and lines to connect various portions of the overall electronic device 10, perform various functions of the electronic device 10, and process data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 120, and invoking data stored in the memory 120. Alternatively, the processor 110 may be implemented in hardware in at least one of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), programmable logic array (Programmable Logic Array, PLA). The processor 110 may integrate one or a combination of several of a central processing unit (Central Processing Unit, CPU), an image processor (Graphics Processing Unit, GPU), and a modem, etc. The CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for being responsible for rendering and drawing of display content; the modem is used to handle wireless communications. It will be appreciated that the modem may not be integrated into the processor 110 and may be implemented solely by a single communication chip.

The Memory 120 may include a random access Memory (Random Access Memory, RAM) or a Read-Only Memory (Read-Only Memory). Memory 120 may be used to store instructions, programs, code, sets of codes, or sets of instructions. The memory 120 may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the various method embodiments described below, etc. The storage data area may also store data created by the electronic device 10 in use (e.g., phonebook, audiovisual data, chat log data), and the like.

Referring to fig. 18, a block diagram of a computer readable storage medium according to an embodiment of the present application is shown. The computer readable medium 1800 has stored therein program code that can be invoked by a processor to perform the methods described in the method embodiments described above.

The computer-readable storage medium 1800 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 1800 includes a non-volatile computer readable medium (non-transitory computer-readable storage medium). The computer readable storage medium 1800 has storage space for program code 1810 that performs any of the method steps described above. The program code can be read from or written to one or more computer program products. Program code 1810 may be compressed, for example, in a suitable form.

In addition, in the embodiment of the present application, the data related to the user needs to be acquired and licensed by the user, and the collection, use, processing and storage of the information need to meet the requirements of the region where the information is located.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application, and are not limiting; although the application has been described in detail with reference to the foregoing embodiments, it will be appreciated by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not drive the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of data processing, comprising:

acquiring a first score determined by an interest heuristic model to be trained according to the similarity between user characteristic data and a preset interest label, wherein the first score is used for representing the possibility that sample data belongs to unknown interests of a user, the sample data is exposure data corresponding to the user, and the exposure data is processed by the user in a preset time period;

Acquiring the intimacy degree between the user and each friend and the interest information of each friend;

acquiring a weight value determined by a joint probability model according to the intimacy degree of each friend and the interest degree of each friend in the sample data, wherein the weight value is used for representing the possibility that the sample is suitable for training the interest heuristic model, and the interest degree of each friend in the sample data is determined based on the known interest information of each friend; the greater the intimacy and the greater the interestingness, the smaller the weight value determined;

adjusting the first score according to the weight value, wherein the weight value is positively correlated with the first score;

training the interest heuristic model to be trained according to the adjusted first score, wherein the trained interest heuristic model is used for determining estimated interest information of a user according to the user characteristic data, the estimated interest information is unknown interest information of the user, and the estimated interest information is used for determining content to be pushed to a client corresponding to the user.

2. The method as recited in claim 1, further comprising:

acquiring a second score determined by the accurate interest model according to the known interest information of the user, wherein the second score is used for representing the possibility that the sample data belong to the known interest of the user, and the known interest information is the determined interest of the user;

Adjusting the weight value according to the first score and the second score, wherein the weight value is used as an output value of a loss function of the joint probability model;

and training the joint probability model according to the adjusted weight value.

3. The method of claim 1, wherein prior to obtaining the first score determined by the heuristic model of interest to be trained based on the user characteristic data, further comprising:

and pre-training the initial interest heuristic model for a designated number of times according to the sample data and the user characteristic data to obtain the interest heuristic model to be trained.

4. A method according to any of claims 1-3, wherein the user characteristic data comprises basic characteristics of the user and behavior data generated by the user for operation of a specific application module of the client and other application modules associated with the specific application module and belonging to the client.

5. A method according to any one of claims 1-3, wherein after training the heuristic model of interest to be trained based on the adjusted first score, further comprising:

Obtaining estimated interest information of a user according to a trained interest heuristic model;

determining content to be pushed according to the estimated interest information of the user;

and pushing the content to be pushed to the client corresponding to the user.

6. The method of claim 5, wherein the determining content to be pushed based on the estimated interest information of the user comprises:

determining the content meeting the specified requirement as the alternative content;

and determining the content to be pushed from the alternative content according to the estimated interest information of the user.

7. The method of claim 6, wherein the determining content meeting the specified requirement comprises, as alternative content:

determining, from among the plurality of contents, content matching the known interest information of the user;

and taking the content except the matching content as the candidate content in the plurality of contents.

8. A data processing apparatus, comprising:

the first acquisition unit is used for acquiring a first score determined by the interest heuristic model to be trained according to the similarity between the user characteristic data and the preset interest tag, wherein the first score is used for representing the possibility that sample data belongs to the unknown interest of the user, the sample data is exposure data corresponding to the user, and the exposure data is processed by the user in a preset time period;

The second acquisition unit is used for acquiring the intimacy degree between the user and each friend and the interest information of each friend; acquiring a weight value determined by a joint probability model according to the intimacy degree of each friend and the interest degree of each friend in the sample data, wherein the weight value is used for representing the possibility that the sample is suitable for training the interest heuristic model, and the interest degree of each friend in the sample data is determined based on the known interest information of each friend; the greater the intimacy and the greater the interestingness, the smaller the weight value determined;

the adjusting unit is used for adjusting the first score according to the weight value, and the weight value is positively related to the first score;

the training unit is used for training the interest heuristic model to be trained according to the adjusted first score, the trained interest heuristic model is used for determining estimated interest information of a user according to the user characteristic data, the estimated interest information is unknown interest information of the user, and the estimated interest information is used for determining content to be pushed to a client corresponding to the user.

9. A computer readable medium, characterized in that the readable medium stores a program code executable by a processor, which program code, when executed by the processor, causes the processor to perform the method of any of claims 1-7.