A data recommendation method, device, computer, and readable storage medium

Abstract

An embodiment of the present application discloses a data recommendation method, apparatus, computer, and readable storage medium, which relates to the field of artificial intelligence. The method includes: obtaining historical login status information of a target user, and predicting the predicted login status of the target user within a target time period according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online state, searching for target user features associated with the target user from a user feature library, and obtaining object features corresponding to at least two candidate objects respectively; performing an association prediction on the object features corresponding to the at least two candidate objects respectively and the target user features to obtain the predicted association degrees between the at least two candidate objects and the target user respectively, and based on the predicted association degrees, determining a target recommended object of the target user from the at least two candidate objects. By adopting the present application, the efficiency of data recommendation can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technologies, and in particular, to a data recommendation method, apparatus, computer, and readable storage medium. Background Art

[0002] With the increasing development of the Internet industry, the types and quantities of application programs are increasing, and the number of users using application programs is also increasing. To better serve users, application programs generally provide customized services for users, that is, recommend customized objects to users. Among them, the current object recommendation generally filters candidate objects through a recall service, and at the same time requests a user feature service, an object feature service, a model service, etc. to score the filtered candidate objects to determine the objects to be recommended to the logged-in user. The process is relatively complex, bringing high computing and storage costs to the online layer and reducing the efficiency of data recommendation. At the same time, filtering candidate objects based on the recall service reduces the service pressure on the online layer, but filtering candidate objects will filter out a large amount of data, and the recall service generally recalls candidate objects based on user features. Therefore, it may cause the objects that the user really likes to be filtered out, reducing the accuracy of the model. Summary of the Invention

[0003] Embodiments of the present application provide a data recommendation method, apparatus, computer, and readable storage medium, which can improve the efficiency of data recommendation.

[0004] On the one hand, an embodiment of the present application provides a data recommendation method, and the method includes:

[0005] Obtain historical login status information of a target user, and predict a predicted login status of the target user within a target time period according to the historical login status information;

[0006] If the predicted login status of the target user within the target time period belongs to an online status, search for target user features associated with the target user in a user feature library, and obtain object features corresponding to at least two candidate objects respectively;

[0007] Perform an association prediction on the object features corresponding to at least two candidate objects respectively and the target user features to obtain a predicted association degree between each of the at least two candidate objects and the target user, and based on the predicted association degree, determine a target recommended object of the target user from the at least two candidate objects; the target recommended object is used as the recommended content of the target user when it is detected that the actual login status of the target user within the target time period is an online status.

[0008] Among them, obtaining historical login status information of a target user and predicting a predicted login status of the target user within a target time period according to the historical login status information includes:

[0009] Obtain the historical login status information of the target user, and obtain at least two default distribution states;

[0010] According to the historical login status information and at least two default distribution states, obtain a state transition matrix;

[0011] Predict the predicted login status of the target user within the target time period according to the state transition matrix.

[0012] Among them, obtaining a state transition matrix according to the historical login status information and at least two default distribution states includes:

[0013] Determine the state change sequence according to the historical login status information, use at least two default distribution states as state nodes respectively, determine the edges between at least two state nodes based on the state change sequence, and generate a state chain;

[0014] Determine the state transition matrix based on the state chain.

[0015] Among them, determining the state transition matrix based on the state chain includes:

[0016] Determine the matrix dimension characteristics based on at least two state nodes in the state chain, determine the matrix parameters according to the edges between at least two state nodes in the state chain, and form an initial state transition matrix with the matrix dimension characteristics and the matrix parameters;

[0017] Perform a transfer process on the initial state transition matrix to obtain the state transition matrix.

[0018] Among them, performing an association prediction on the object characteristics corresponding to at least two candidate objects and the target user characteristics respectively to obtain the predicted association degrees between at least two candidate objects and the target user respectively includes:

[0019] Perform feature splicing on the object characteristics of the i-th candidate object among at least two candidate objects and the target user characteristics to obtain the i-th spliced feature; i is a positive integer, and i is less than or equal to the candidate quantity of at least two candidate objects;

[0020] Perform an association prediction on the i-th spliced feature based on the data recommendation model to obtain the predicted association degree between the i-th candidate object and the target user until the predicted association degrees between at least two candidate objects and the target user are obtained respectively.

[0021] Among them, performing an association prediction on the i-th spliced feature based on the data recommendation model to obtain the predicted association degree between the i-th candidate object and the target user includes:

[0022] Perform vector conversion on at least two sub-features in the i-th spliced feature based on the data recommendation model to obtain conversion vectors corresponding to at least two sub-features respectively;

[0023] Perform feature fusion on at least two sub - features in the \(i\) - th splicing feature to obtain the \(i\) - th first fusion feature, perform feature inner - product processing on the transformation vectors respectively corresponding to the at least two sub - features to obtain the \(i\) - th second fusion feature, and perform activation processing on the transformation vectors respectively corresponding to the at least two sub - features to obtain the \(i\) - th third fusion feature;

[0024] Perform correlation prediction on the \(i\) - th first fusion feature, the \(i\) - th second fusion feature, and the \(i\) - th third fusion feature to obtain the predicted correlation degree between the \(i\) - th candidate object and the target user.

[0025] Among them, based on the predicted correlation degree, determining the target recommended object of the target user from at least two candidate objects includes:

[0026] Obtain a recommended quantity threshold, perform correlation sorting on at least two candidate objects based on the predicted correlation degree, and obtain \(N\) candidate objects from the at least two candidate objects after correlation sorting as the target recommended objects of the target user; \(N\) is a positive integer, and \(N\) is greater than or equal to the recommended quantity threshold.

[0027] Among them, the target user and the target recommended object are stored in the offline layer; the method further includes:

[0028] Send the target user and the target recommended objects of the target user to the online layer based on the offline layer, and perform associated storage of the target user and the target recommended objects in the user recommendation library in the online layer;

[0029] In the online layer, in response to a login operation for the target user, obtain the target recommended objects associated with the target user from the user recommendation library based on the associated storage relationship, and output the target recommended objects.

[0030] Among them, the method further includes:

[0031] In the online layer, in response to a login operation for the to - be - processed user, search for the to - be - output recommended objects associated with the to - be - processed user from the user recommendation library based on the associated storage relationship;

[0032] If there are no to - be - output recommended objects associated with the to - be - processed user in the user recommendation library, obtain default recommended objects and output the default recommended objects for the to - be - processed user.

[0033] Among them, the method further includes:

[0034] Obtain at least two training user samples, and obtain the sample user features respectively associated with the at least two training user samples from the user feature library;

[0035] Obtain the sample object features associated with each training user sample, and perform model training based on the sample user features and sample object features corresponding to at least two training user samples to generate a data recommendation model.

[0036] Among them, obtaining at least two training user samples includes:

[0037] During the training period, obtain at least two candidate users associated with the application; the application is used to output the target recommendation object of the target user;

[0038] Obtain the candidate login status corresponding to at least two candidate users during the training period, and determine the candidate users with the candidate login status including the online status as training user samples;

[0039] Obtaining the sample object features associated with each training user sample includes:

[0040] Count the historical associated objects generated by the training user samples in the online state, and determine the historical object features of the historical associated objects as the sample object features associated with the training user samples.

[0041] Among them, the user feature library includes at least two statistical users and the user features of each statistical user; the at least two statistical users include statistical user j; j is a positive integer, and j is less than or equal to the number of users of the at least two statistical users;

[0042] Obtain the sample user features respectively associated with at least two training user samples from the user feature library, including:

[0043] Obtain the training sample hash values corresponding to at least two training user samples respectively, and generate a sample matching bit array according to the training sample hash values corresponding to at least two training user samples respectively;

[0044] Generate the user hash value of statistical user j, and obtain the user status value of statistical user j from the sample matching bit array according to the user hash value of statistical user j;

[0045] If there is at least one missing status value in the user status value of statistical user j, it is determined that statistical user j is not associated with at least two training user samples. If the user status values of statistical user j are all valid status values, it is determined that statistical user j is associated with at least two training user samples;

[0046] Determine the user features corresponding to the statistical users associated with at least two training user samples in the user feature library as the sample user features associated with at least two training user samples.

[0047] Among them, the at least two training user samples include training user sample k; k is a positive integer, and k is less than or equal to the number of samples of the at least two training user samples;

[0048] Training a model based on the sample user features and sample object features respectively corresponding to at least two training user samples to generate a data recommendation model, including:

[0049] Predicting the sample user features and sample object features of the training user sample k based on the initial data recommendation model to obtain the sample prediction correlation degree between the training user sample k and the sample object features of the training user sample k;

[0050] Training the initial data recommendation model based on the sample prediction correlation degree and the model loss function to generate a data recommendation model.

[0051] One aspect of the embodiments of the present application provides a data recommendation device, and the device includes:

[0052] A status prediction module, configured to obtain the historical login status information of the target user, and predict the predicted login status of the target user within the target time period according to the historical login status information;

[0053] A feature acquisition module, configured to, if the predicted login status of the target user within the target time period belongs to the online status, search for the target user features associated with the target user from the user feature library, and obtain the object features respectively corresponding to at least two candidate objects;

[0054] A recommendation prediction module, configured to perform an association prediction on the object features respectively corresponding to at least two candidate objects and the target user features to obtain the prediction correlation degrees between at least two candidate objects and the target user respectively, and determine the target recommendation object of the target user from at least two candidate objects based on the prediction correlation degrees; the target recommendation object is used as the recommended content of the target user when it is detected that the actual login status of the target user within the target time period is the online status.

[0055] Wherein, the status prediction module includes:

[0056] A distribution acquisition unit, configured to obtain the historical login status information of the target user and obtain at least two default distribution states;

[0057] A matrix acquisition unit, configured to obtain a state transition matrix according to the historical login status information and at least two default distribution states;

[0058] A status prediction unit, configured to predict the predicted login status of the target user within the target time period according to the state transition matrix.

[0059] Wherein, the matrix acquisition unit includes:

[0060] A spectrum generation subunit, configured to determine a state change sequence according to historical login status information, use at least two default distribution states as state nodes respectively, determine the edges between at least two state nodes based on the state change sequence, and generate a state chain;

[0061] A matrix determination subunit, configured to determine a state transition matrix based on the state chain.

[0062] Among them, the matrix determination subunit is specifically configured to:

[0063] Determine matrix dimension features based on at least two state nodes in the state chain, determine matrix parameters according to the edges between at least two state nodes in the state chain, and form an initial state transition matrix with the matrix dimension features and matrix parameters;

[0064] Perform a transfer process on the initial state transition matrix to obtain a state transition matrix.

[0065] Among them, in terms of performing an association prediction on the object features corresponding to at least two candidate objects and the target user features respectively to obtain the prediction association degrees between at least two candidate objects and the target user respectively, this recommendation prediction module includes:

[0066] A feature splicing unit, configured to splice the object feature of the i-th candidate object among at least two candidate objects and the target user feature to obtain the i-th spliced feature; i is a positive integer, and i is less than or equal to the number of candidates of at least two candidate objects;

[0067] An association prediction unit, configured to perform an association prediction on the i-th spliced feature based on the data recommendation model to obtain the prediction association degree between the i-th candidate object and the target user, until the prediction association degrees between at least two candidate objects and the target user are obtained respectively.

[0068] Among them, this association prediction unit includes:

[0069] A vector conversion subunit, configured to perform vector conversion on at least two sub-features in the i-th spliced feature based on the data recommendation model to obtain conversion vectors corresponding to at least two sub-features respectively;

[0070] A feature fusion subunit, configured to perform feature fusion on at least two sub-features in the i-th spliced feature to obtain the i-th first fusion feature, perform feature inner product processing on the conversion vectors corresponding to at least two sub-features respectively to obtain the i-th second fusion feature, and perform activation processing on the conversion vectors corresponding to at least two sub-features respectively to obtain the i-th third fusion feature;

[0071] A feature prediction subunit, configured to perform an association prediction on the i-th first fusion feature, the i-th second fusion feature, and the i-th third fusion feature to obtain the prediction association degree between the i-th candidate object and the target user.

[0072] Among them, in terms of determining the target recommended object of the target user from at least two candidate objects based on the prediction correlation degree, the recommendation prediction module is specifically configured to:

[0073] Obtain a recommended quantity threshold, perform correlation ranking on at least two candidate objects based on the prediction correlation degree, and obtain N candidate objects from the at least two candidate objects after correlation ranking as the target recommended objects of the target user; N is a positive integer, and N is greater than or equal to the recommended quantity threshold.

[0074] Among them, the target user and the target recommended object are stored in the offline layer; the device further includes:

[0075] A recommendation storage module, configured to send the target user and the target recommended objects of the target user to the online layer based on the offline layer, and perform associated storage of the target user and the target recommended objects in the user recommendation library of the online layer;

[0076] A recommendation output module, configured to, in the online layer, in response to a login operation for the target user, obtain the target recommended objects associated with the target user from the user recommendation library based on the associated storage relationship, and output the target recommended objects.

[0077] Among them, the device further includes:

[0078] An object search module, configured to, in the online layer, in response to a login operation for the to-be-processed user, search for the to-be-output recommended objects associated with the to-be-processed user from the user recommendation library based on the associated storage relationship;

[0079] A default output module, configured to, if there are no to-be-output recommended objects associated with the to-be-processed user in the user recommendation library, obtain default recommended objects and output the default recommended objects for the to-be-processed user.

[0080] Among them, the device further includes:

[0081] A sample acquisition module, configured to acquire at least two training user samples, and acquire sample user features respectively associated with the at least two training user samples from the user feature library;

[0082] A model training module, configured to acquire the sample object features associated with each training user sample, and perform model training based on the sample user features and sample object features respectively corresponding to the at least two training user samples to generate a data recommendation model.

[0083] Among them, in terms of acquiring at least two training user samples, the sample acquisition module includes:

[0084] A candidate acquisition unit, configured to acquire at least two candidate users associated with an application during a training period; the application is used to output target recommendation objects for a target user.

[0085] A sample selection unit, configured to acquire candidate login statuses corresponding to at least two candidate users during the training period, and determine candidate users with an online status in the candidate login statuses as training user samples.

[0086] In terms of acquiring sample object features associated with each training user sample, the model training module includes:

[0087] An object acquisition unit, configured to count historical associated objects generated by a training user sample in an online state, and determine historical object features of the historical associated objects as sample object features associated with the training user sample.

[0088] Wherein, the user feature library includes at least two statistical users and user features of each statistical user; the at least two statistical users include a statistical user j; j is a positive integer, and j is less than or equal to the number of users of the at least two statistical users.

[0089] In terms of acquiring sample user features respectively associated with at least two training user samples from the user feature library, the sample acquisition module includes:

[0090] An array generation unit, configured to acquire training sample hash values corresponding to at least two training user samples respectively, and generate a sample matching bit array according to the training sample hash values corresponding to the at least two training user samples respectively.

[0091] A hash matching unit, configured to generate a user hash value of the statistical user j, and acquire a user status value of the statistical user j from the sample matching bit array according to the user hash value of the statistical user j.

[0092] An association determination unit, configured to determine that the statistical user j is not associated with at least two training user samples if there is at least one missing status value in the user status value of the statistical user j, and determine that the statistical user j is associated with at least two training user samples if all user status values of the statistical user j are valid status values.

[0093] A feature determination unit, configured to determine user features corresponding to statistical users associated with at least two training user samples in the user feature library as sample user features associated with at least two training user samples.

[0094] Wherein, the at least two training user samples include a training user sample k; k is a positive integer, and k is less than or equal to the number of samples of the at least two training user samples.

[0095] In terms of training a model to generate a data recommendation model based on the sample user features and sample object features corresponding to at least two training user samples respectively, the model training module includes:

[0096] A sample prediction unit, configured to predict the sample user features and sample object features of the training user sample k based on the initial data recommendation model, and obtain the sample prediction correlation degree between the training user sample k and the sample object features of the training user sample k;

[0097] A model training unit, configured to train the initial data recommendation model based on the sample prediction correlation degree and the model loss function to generate a data recommendation model.

[0098] On the one hand, an embodiment of the present application provides a computer device, including a processor, a memory, and an input / output interface;

[0099] The processor is respectively connected to the memory and the input / output interface. Among them, the input / output interface is used to receive and output data, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device including the processor executes the data recommendation method in one aspect of the embodiment of the present application.

[0100] On the one hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and the computer program is adapted to be loaded and executed by a processor so that a computer device having the processor executes the data recommendation method in one aspect of the embodiment of the present application.

[0101] On the one hand, an embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device executes the methods provided in various alternative manners in one aspect of the embodiment of the present application.

[0102] Implementing the embodiments of the present application will have the following beneficial effects:

[0103] In an embodiment of the present application, a computer device may obtain historical login status information of a target user, predict the predicted login status of the target user within a target time period according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online status, search for target user features associated with the target user in a user feature library, and obtain object features corresponding to at least two candidate objects respectively; perform an association prediction on the object features corresponding to at least two candidate objects respectively and the target user features to obtain the predicted association degrees between at least two candidate objects and the target user respectively, and based on the predicted association degrees, determine a target recommended object of the target user from at least two candidate objects; the target recommended object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status. By predicting the login status of the target user, the predicted login status of the target user is obtained. When the predicted login status belongs to the online status, the target recommended object of the target user can be determined, so that the target recommended object can be determined in advance, and when the target user logs in, the target recommended object predicted in advance can be directly obtained, thereby improving the efficiency of data recommendation. Further, in the present application, the association degrees between at least two candidate objects and the target user are predicted respectively, and full-object prediction scoring (i.e., predicted association degree) is realized, improving the accuracy of data recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0105] Figure 1 FIG. is a network interaction architecture diagram of a data recommendation provided by an embodiment of the present application;

[0106] Figure 2 FIG. is a scenario schematic diagram of a data recommendation provided by an embodiment of the present application;

[0107] Figure 3 FIG. is a flowchart of a method for data recommendation provided by an embodiment of the present application;

[0108] Figure 4 FIG. is a scenario schematic diagram of a transition matrix generation provided by an embodiment of the present application;

[0109] Figure 5 FIG. is a network schematic diagram of a data recommendation model provided by an embodiment of the present application;

[0110] Figure 6It is a schematic diagram of the training process of a data recommendation model provided by an embodiment of the present application;

[0111] Figure 7 It is a schematic diagram of the generation and filtering scenario of a sample matching bit array provided by an embodiment of the present application;

[0112] Figure 8 It is a schematic diagram of a data recommendation architecture provided by an embodiment of the present application;

[0113] Figure 9 It is another schematic diagram of a data recommendation architecture provided by an embodiment of the present application;

[0114] Figure 10 It is a schematic diagram of a data recommendation device provided by an embodiment of the present application;

[0115] Figure 11 It is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0116] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0117] The use of user data in the present application complies with the relevant regulations of laws and regulations.

[0118] Among them, in the embodiments of the present application, the computer device can batch predict the status of users to obtain the predicted login status corresponding to at least two users respectively, and then determine the recommended object corresponding to each user based on the predicted login status corresponding to each user. That is to say, in the embodiments of the present application, there may be a large number of users who need to be recommended and predicted, and there may also be a large number of candidate objects for determining the recommended objects of users. That is, a large amount of data may be generated in the present application, and a large amount of data needs to be stored and processed, etc. Therefore, big data technology can be used to realize the prediction of the predicted login status of each user, and based on the predicted login status, determine the recommended object of each user from at least two candidate objects, thereby improving the efficiency of data recommendation. Among them, at least two mentioned in the present application all refer to two or more. For example, at least two users refer to two or more users, and at least two candidate objects refer to two or more candidate objects, etc.

[0119] Among them, big data refers to a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time range. It is a massive, high-growth, and diverse information asset that requires new processing models to have stronger decision-making power, insight discovery ability, and process optimization ability. With the advent of the cloud era, big data has also attracted more and more attention. Big data requires special technologies to effectively process a large amount of data tolerated over time. Technologies applicable to big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems.

[0120] Furthermore, the computer device can use technologies such as machine learning in the field of artificial intelligence to perform correlation prediction on the object features corresponding to at least two candidate objects and the target user features, so as to obtain the predicted association degrees between at least two candidate objects and the target user respectively, and then determine the target recommended object of the target user according to the predicted association degrees.

[0121] Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, a theory, method, technology, and application system that perceives the environment, acquires knowledge, and uses knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence, so as to process the characteristics of multimedia data on each data channel and at each pixel point, and make the processing results as similar as possible to the quality assessment results of human intelligence for multimedia data. Artificial intelligence is also to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning, and decision-making.

[0122] Among them, this application mainly involves directions such as machine learning / deep learning (such as data recommendation models, etc.). A data recommendation model can be obtained through learning to predict the predicted association degrees between at least two candidate objects and the target user respectively.

[0123] Deep learning (DL) is a new research direction in the field of machine learning (ML). Deep learning is to learn the internal laws and representation levels of sample data. Deep learning usually includes technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0124] In the embodiments of this application, please refer to Figure 1 , Figure 1It is a network interaction architecture diagram for data recommendation provided by an embodiment of the present application. As Figure 1 shown, data interaction can occur between computer device 101 and user devices (such as user device 102a, user device 102b, user device 102c, etc.). The computer device can predict a target user and determine a target recommended object for the target user from at least two candidate objects. When the target user logs in, computer device 101 can send the target recommended object to the target user terminal where the target user is located. For example, computer device 101 obtains at least two users, takes each of the at least two users as the target user in turn to traverse the at least two users, determines the predicted login status of each user, and determines the recommended object for each user based on the predicted login status. Specifically, assuming that the at least two users include M users, where M is a positive integer, taking the first user as the target user, predicting the predicted login status of the first user. If the predicted login status of the first user belongs to the online state, then predicting the predicted association degree between each of the at least two candidate objects and the first user, and determining the target recommended object for the first user based on the predicted association degree; taking the second user as the target user, predicting the predicted login status of the second user. If the predicted login status of the second user belongs to the online state, then predicting the predicted association degree between each of the at least two candidate objects and the second user, and determining the target recommended object for the second user based on the predicted association degree;...; taking the Mth user as the target user, predicting the predicted login status of the Mth user. If the predicted login status of the Mth user belongs to the online state, then predicting the predicted association degree between each of the at least two candidate objects and the Mth user, and determining the target recommended object for the Mth user based on the predicted association degree. Optionally, the users among the at least two users whose predicted login status belongs to the online state and their target recommended objects can be stored, denoted as the object recommendation list. Further, assuming that computer device 101 monitors the login of the second user, in response to the login operation for the second user, obtaining the target recommended object of the second user from the object recommendation list. Assuming that the user device where the second user is located is user device 102b, then computer device 101 pushes the target recommended object of the second user to user device 102b so that user device 102b can output the target recommended object of the second user. Through the above process, the predicted login status of each user within the target time period can be predicted, and the target recommended objects of the users whose predicted login status belongs to the online state can be determined, enabling direct data recommendation based on the previously predicted target recommended objects when the user logs in, thereby improving the efficiency of data recommendation. Moreover, when determining the target recommended object, it is determined based on at least two objects, making the prediction of the recommended object more comprehensive, thereby improving the accuracy of data recommendation.

[0125] Specifically, please refer toFigure 2 , Figure 2 is a schematic diagram of a data recommendation scenario provided by an embodiment of the present application. As Figure 2 shown, a computer device can obtain a user cluster 201 associated with an application program. The user cluster 201 includes target users 2011, 2012, 2013, etc. The computer device obtains the historical login status information of the target user 2011, predicts the predicted login status of the target user 2011 within a target time period based on the historical login status information. Assuming that the predicted login status of the target user 2011 belongs to the online state, the computer device obtains the target user characteristics of the target user 2011, and obtains at least two candidate objects associated with the application program. The at least two candidate objects include all candidate objects associated with the application program. The computer device can obtain the object characteristics 202 corresponding to each candidate object respectively, including the object characteristic 1 of candidate object 1, the object characteristic 2 of candidate object 2, and the object characteristic p of candidate object p, etc., where p is a positive integer and p is the number of candidate objects included in the at least two candidate objects. The computer device associates and predicts the object characteristics 202 corresponding to each candidate object with the target user characteristics to obtain the predicted association degrees 203 between at least two candidate objects and the target user 2011 respectively. Based on the predicted association degrees 203, the computer device determines the target recommended object of the target user 2011 from the at least two candidate objects, where the number of the target recommended objects of the target user 2011 is at least one (i.e., one or at least two). Specifically, the computer device associates and predicts the object characteristic 1 with the target user characteristics to obtain the predicted association degree 1 between candidate object 1 and the target user 2011; associates and predicts the object characteristic 2 with the target user characteristics to obtain the predicted association degree 2 between candidate object 2 and the target user 2011;...; associates and predicts the object characteristic p with the target user characteristics to obtain the predicted association degree p between candidate object p and the target user 2011. The computer device determines the target recommended object of the target user 2011 from the at least two candidate objects according to the predicted association degrees 1, predicted association degrees 2,..., and predicted association degree p.

[0126] The computer device obtains the historical login status information of the target user 2013, predicts the predicted login status of the target user 2013 within the target time period based on the historical login status information. Assuming that the predicted login status of the target user 2013 belongs to the offline status, which means that the target user 2013 is unlikely to log in to the application within the target event segment, then the computer device does not predict the target recommended object of the target user 2013. Or, the computer device obtains the default recommended object and uses the default recommended object as the target recommended object of the target user 2013. Among them, if the computer device does not predict the target recommended object of the target user 2013, it can directly obtain the default recommended object when the target user 2013 logs in during the target time period and push the default recommended object to the user terminal where the target user 2013 is located. Since there may be more than one user with a predicted login status of offline, the time and resources consumed for predicting the target recommended object can be reduced, and the default recommended object only needs to be stored once, saving storage space. If the default recommended object is used as the target recommended object of the target user 2013 in advance, the computer device can directly find the target recommended object of the target user 2013 when the target user 2013 logs in during the target time period, without having to obtain the default recommended object when the target recommended object of the target user 2013 cannot be found, thus reducing the time consumed for the target user 2013 to log in. The above two processing methods can both be implemented in this application. In practical applications, the determination method of the target recommended object of the target user 2013 can be selected according to needs, and no restrictions are imposed here. Similarly, for the target users in the user cluster 201 whose predicted login status belongs to the offline status, the computer device can take the determination process of the target recommended object of the target user 2013 as an example to determine the target recommended object of the target users whose predicted login status belongs to the offline status.

[0127] Similarly, based on the determination processes of the target recommended objects of the target user 2011 and the target user 2013, the target recommended objects of all the target users included in the user cluster 201 can be determined, and no further elaboration will be given here.

[0128] It can be understood that the computer devices and user devices mentioned in the embodiments of the present application include, but are not limited to, terminal devices or servers. In other words, the computer device can be a server or a terminal device, or a system composed of a server and a terminal device; the user device can be a server or a terminal device, or a system composed of a server and a terminal device. Among them, the above-mentioned terminal device can be an electronic device, including but not limited to mobile phones, tablet computers, desktop computers, laptop computers, palmtop computers, in-vehicle devices, augmented reality / virtual reality (AR / VR) devices, head-mounted displays, smart TVs, wearable devices, smart speakers, digital cameras, cameras, and other mobile internet devices (MIDs) with network access capabilities. Among them, the above-mentioned server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, vehicle-road coordination, content delivery network (CDN), and big data and artificial intelligence platforms.

[0129] Optionally, the data involved in the embodiments of the present application can be stored in a computer device, or the data can be stored based on cloud storage technology, which is not limited here. Among them, cloud storage is a new concept extended and developed on the basis of the cloud computing concept. A distributed cloud storage system (hereinafter referred to as a storage system) refers to a storage system that combines a large number of various types of storage devices (storage devices are also called storage nodes) in the network through cluster applications, grid technology, and distributed file systems, and works together through application software or application interfaces to jointly provide data storage and service access functions to the outside world. In other words, the data involved in the embodiments of the present application can be stored based on a distributed cloud storage system.

[0130] Currently, the storage method of the storage system is as follows: Create a logical volume. When creating a logical volume, physical storage space is allocated for each logical volume. This physical storage space may be a certain storage device or the disks of several storage devices. The client stores data on a certain logical volume, that is, stores the data on the file system. The file system divides the data into many parts, and each part is a storage object. The storage object contains not only the data but also additional information such as the data identifier (ID, IDentity). The file system writes each object into the physical storage space of the logical volume respectively, and the file system will record the storage location information of each storage object. Thus, when the computer device requests to access the data, the file system can enable the computer device to access the data according to the storage location information of each system object. For example, when the computer device requests to access the user feature library, the file system can enable the computer device to access the user feature library according to the storage location information of the user feature library.

[0131] The process of the storage system allocating physical storage space for the logical volume is specifically as follows: According to the capacity estimation of the objects stored in the logical volume (this estimation often has a large margin relative to the capacity of the objects to be actually stored) and the group of the redundant array of independent disks (RAID), the physical storage space is pre-divided into stripes. A logical volume can be understood as a stripe, thereby allocating physical storage space for the logical volume. Through cloud storage technology, the storage pressure of the computer device itself can be reduced. When the computer device implements the embodiments of the present application, what is implemented is the preprocessing process of data recommendation for the target user. Therefore, the time loss that may be brought when the computer device obtains data based on cloud storage technology has no impact on the actual login process of the target user and does not affect the use of the application program. Among them, the computer device can also be stored in the offline layer of the computer device. Since the offline layer has a large storage space, the computer device can also be directly stored in this offline layer to reduce the time loss caused by data interaction and improve the efficiency of preprocessing data recommendation. Therefore, the data involved in the present application can be stored in the computer device or stored based on cloud storage technology for this data. The specific storage method can be determined based on needs.

[0132] Further, please refer to Figure 3 , Figure 3 which is a flowchart of a data recommendation method provided by an embodiment of the present application. As Figure 3 shown, the data recommendation process includes the following steps:

[0133] Step S301, obtain the historical login status information of the target user, and predict the predicted login status of the target user within the target time period according to the historical login status information.

[0134] In an embodiment of the present application, a computer device may obtain historical login status information of a target user, predict the login status of the target user within a target time period based on the historical login status information, and obtain a predicted login status of the target user within the target time period. Specifically, the computer device may obtain the historical login status generated by the target user within a status prediction period, determine the historical login status information of the target user based on the historical login status generated within the status prediction period. The historical login status information may include one or at least two historical login statuses, and the one or at least two historical login statuses form a status change sequence. Among them, the status prediction period is used to represent the duration of the time period that needs to be counted when counting the historical login status information of the target user. For example, assuming that the status prediction period is seven days, the historical login status of the target user within seven days is obtained to form the historical login status information. Among them, the historical login status belongs to at least two default distribution states. The historical login status information is used to represent the change of the historical login status of the target user within the status prediction period. For example, assuming that the at least two default distribution states include "application installation status, message receiving status, page access status, etc.", the historical login status of the target user on the first day within the status prediction period is "message receiving status", the historical login status on the second day within the status prediction period is "page access status", the historical login status on the third day within the status prediction period is "message receiving status",..., according to the historical login status of the target user within the status prediction period, the historical login status information "message receiving status -> page access status -> message receiving status..." is obtained, and the predicted login status of the target user is determined based on the historical login status information. Among them, the at least two default distribution states may also be composed of other default distribution states, such as including "application click status, application usage status, application idle status, etc.", or including "application usage status and application idle status, etc.", and so on. That is, the default distribution states included in the at least two default distribution states can be set as needed and are not limited here. The computer device may obtain the historical login status of the target user within the status prediction period based on the at least two default distribution states and determine the historical login status information.

[0135] Among them, the computer device can obtain the historical login status information of the target user and obtain at least two default distribution states; based on the historical login status information and at least two default distribution states, obtain a state transition matrix; and predict the predicted login status of the target user within the target time period according to the state transition matrix. Among them, the various historical login states in the historical login status information of the target user are independent of each other, that is, given the current historical login state of the target user, the historical login states of the target user in the past and in the future are independent of each other, and the probabilities between the historical login states in the historical login information satisfy Pr(H1 = h1, H2 = h2, …, H t = h t ) > 0, where the sequence (H1, H2, …, H t ) is used to represent the variable sequence of the historical login states generated by the target user within the state prediction period, and the sequence (h1, h2, …, h t ) is used to represent the actual historical login states of the target user within the state prediction period, that is, (H1 = h1, H2 = h2, …, H t = h t ) can be used to represent the historical login status information of the target user within the state prediction period. Among them, the various historical login states in the historical login status information are independent of each other. Therefore, when predicting the predicted login status of the target user within the target time period based on the historical login status information, the characteristics of the prediction probability can be as shown in formula ①:

[0136]

[0137] Among them, as shown in formula ①, the probability of predicting the predicted login status of the target user under the historical login status information is the same as the probability of predicting the predicted login status of the target user under the adjacent historical login status, that is, given (H1, H2, …, H t ) to predict the probability of H t+1 is the same as the probability of predicting the probability of H t to predict H t+1 . Among them, the countable set S composed of the possible values of H can be called the "state space", and the "state space" includes at least two default distribution states.

[0138] Furthermore, the computer device can determine the state change sequence according to the historical login status information, that is, (h1, h2, …, h t), at least two default distribution states are respectively used as state nodes, and edges between at least two state nodes are determined based on the state change sequence to generate a state chain; based on the state chain, a state transition matrix is determined. Based on the characteristics of the prediction probability, the computer device can determine the predicted login state of the target user within the target time period according to the state transition matrix. Among them, the state chain includes multiple state nodes, and the state chain is a chain structure composed of the transition probabilities between the multiple state nodes, such as a Markov chain, etc., which is not limited here. Among them, the computer device can determine the matrix dimension feature based on at least two state nodes in the state chain, determine the matrix parameter according to the edges between at least two state nodes in the state chain, form an initial state transition matrix with the matrix dimension feature and the matrix parameter, and perform a transition process on the initial state transition matrix to obtain the state transition matrix. Among them, the transition process of the state transition matrix can be as shown in formula ②:

[0139]

[0140] For example, please refer to Figure 4 , Figure 4 which is a schematic diagram of a transfer matrix generation scenario provided by an embodiment of the present application. As Figure 4 shown, the computer device uses at least two default distribution states 401 as state nodes 402 respectively. Assuming that the at least two default distribution states 401 include default distribution state S1, default distribution state S2, default distribution state S3, and default distribution state S4, state nodes S1, S2, S3, and S4 are obtained, and edges between each state node are determined based on the historical login state information 403 to generate a state chain 404. Among them, the historical login states included in the historical login state information belong to at least two default distribution states, and the type of the historical login state is less than or equal to the type of the default distribution state. Further, as Figure 4 shown, the computer device can determine the matrix dimension feature 4051 based on at least two state nodes in the state chain 404, determine the matrix parameter 4052 according to the edges between at least two state nodes in the state chain 404, and form an initial state transition matrix 405 with the matrix dimension feature 4051 and the matrix parameter, also known as a one-step state transition matrix, and this one-step state transition matrix can be denoted as Among them, the sub-parameters in the matrix parameter are used to represent the transition probability between the dimensional sub-features corresponding to the row number where the sub-parameter is located and the dimensional sub-features where the column number is located. For example, the sub-parameter "0.6" located in the second row and the third column of the matrix parameter is used to represent that the transition probability from the dimensional sub-feature S2 to the dimensional sub-feature S3 is 0.6. Optionally, the rows and columns in the initial state transition matrix 405 can also be swapped, which is not restricted here. Based on this one-step state transition matrix and formula ②, the one-step state transition matrix is subjected to n-step transition processing to obtain a state transition matrix. In other words, the computer device can perform transition processing on the initial state transition matrix 405 based on formula ② to obtain a state transition matrix, so that the state transition matrix can represent the change probability of the target user between each default distribution state, and determine the predicted login state of the target user within the target time period according to the state transition matrix and the historical login state information.

[0141] Optionally, the computer device can also directly predict the historical login state information of the target user based on the state prediction model to obtain the predicted login state of the target user within the target time period.

[0142] Among them, the target time period can be determined according to the object prediction cycle. Specifically, a periodic time period can be obtained based on the object prediction cycle. For example, if the object prediction cycle is one day, assuming that the target user is predicted on December 20th, then the predicted login state of the target user within the target time period is predicted according to the historical login state information of the target user, and the target time period is this day of December 21st.

[0143] Furthermore, the computer device can detect the predicted login state of the target user. If the predicted login state of the target user within the target time period belongs to the offline state, no processing is performed on the target user, or the default recommended object is determined as the target recommended object of the target user; if the predicted login state of the target user within the target time period belongs to the online state, step S302 is executed. Among them, the online state is used to indicate that it is predicted that the target user will use the application program within the target time period, and the offline state is used to indicate that it is predicted that the target user will not use the application program within the target time period.

[0144] Step S302, if the predicted login state of the target user within the target time period belongs to the online state, then search for the target user features associated with the target user in the user feature library, and obtain the object features corresponding to at least two candidate objects.

[0145] In an embodiment of the present application, if the predicted login status of the target user within the target time period belongs to the online state, the target user features associated with the target user are searched from the user feature library. The user feature library includes at least two statistical users associated with the application program and the user features of each statistical user. The computer device can search for the target user features associated with the target user from the user features. The user features may include multiple user attribute features, such as user gender features (e.g., male, female), user age features (e.g., features composed of at least two age groups such as minors), user address features (e.g., the address area or province to which they belong), and user label features (e.g., labels added by the user to the application program), etc. Further, the computer device can obtain candidate objects associated with the application program, and obtain the object features corresponding to each candidate object. The candidate objects may be video objects (e.g., movies, self-made short dramas, or educational videos, etc.), text objects (e.g., novels or text materials, etc.), image objects, or combined objects (e.g., combinations of text objects and image objects), etc. The computer device obtains at least two candidate objects associated with the application program, and the at least two candidate objects refer to all objects that can be obtained. The object features may be generated according to the video type to which the candidate object belongs. For example, the application program is an application program for playing videos, and the application program includes at least two video objects. The computer device can obtain all the video objects associated with the application program, and obtain the object features of each video object. The object features of each video object are the object features corresponding to the at least two candidate objects respectively. Optionally, the object features are stored in the object feature library. The computer device stores the user feature library and the object feature library in the offline layer. The storage space of the offline layer is large and can perform large-scale data processing. Therefore, when the computer device executes step S301 and step S302 in the offline layer, the efficiency of data processing can be improved, and not too much working pressure will be brought to the offline layer.

[0146] Step S303: Perform an association prediction on the object features corresponding to at least two candidate objects and the target user features to obtain the predicted association degrees between at least two candidate objects and the target user respectively. Based on the predicted association degrees, determine the target recommended object of the target user from at least two candidate objects.

[0147] In an embodiment of the present application, the computer device can splice the object features of the i-th candidate object among at least two candidate objects with the target user features to obtain the i-th spliced feature; i is a positive integer, and i is less than or equal to the number of candidates of at least two candidate objects; perform an association prediction on the i-th spliced feature based on the data recommendation model to obtain the predicted association degree between the i-th candidate object and the target user, until the predicted association degrees between at least two candidate objects and the target user are obtained respectively.

[0148] Among them, when the computer device performs an association prediction on the i-th splicing feature to obtain the predicted association degree between the i-th candidate object and the target user, the computer device may perform vector conversion on at least two sub-features in the i-th splicing feature based on the data recommendation model to obtain conversion vectors respectively corresponding to the at least two sub-features; perform feature fusion on the at least two sub-features in the i-th splicing feature to obtain the i-th first fusion feature, perform feature inner product processing on the conversion vectors respectively corresponding to the at least two sub-features to obtain the i-th second fusion feature, and perform activation processing on the conversion vectors respectively corresponding to the at least two sub-features to obtain the i-th third fusion feature; perform an association prediction on the i-th first fusion feature, the i-th second fusion feature, and the i-th third fusion feature to obtain the predicted association degree between the i-th candidate object and the target user.

[0149] Specifically, reference may be made to Figure 5 , Figure 5 which is a network schematic diagram of a data recommendation model provided by an embodiment of the present application. As Figure 5 shown, the data recommendation model may include a Sparse Features Layer, a Dense Embeddings Layer, a Factorization Machine Layer (also referred to as a factorization machine), a Hidden Layer, an Output Units (also referred to as an output unit), etc. The computer device may perform feature splicing on the object feature of the i-th candidate object and the target user feature to obtain the i-th splicing feature 501, and the i-th splicing feature 501 includes at least two sub-features, such as sub-feature 5011, sub-feature 5012, and sub-feature 5013, etc. Optionally, the computer device may directly splice the object feature of the i-th candidate object and the target user feature to obtain the i-th splicing feature (target user feature, object feature of the i-th candidate object); or add the target user feature and the object feature of the i-th candidate object to the default feature to obtain the i-th splicing feature 501.

[0150] Taking the example of adding the target user features and the object features of the i-th candidate object to the default features, assume that the user features include user gender features and user age features, and the object features include object type features and object duration features, etc. Among them, the user gender feature can be represented as (0, 0), representing male, female, etc. respectively; the user age feature can be represented as (0, 0, 0, 0), representing minor, 18 to 30 years old, 31 to 45 years old, and over 46 years old, etc. respectively; the object type feature can be represented as (0, 0, 0, 0), representing entertainment, game, education, short sentence, etc. respectively; the object duration feature can be represented as (0, 0, 0, 0), representing within half an hour, half an hour to one hour, one hour to two hours, and over two hours, etc. respectively. That is, the default feature is (0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0). Assume that the target user gender feature in the target user features is female, the target user age feature is 18 to 30 years old, the object type feature in the object features of the i-th candidate object is education, and the object duration feature is within half an hour. Then add the target user features and the object features of the i-th candidate object to the default features to obtain the i-th concatenated feature (0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0). Among them, the above is an optional way of composing sub-features, and it can also be changed according to needs. That is, the user features can be composed of other user attribute features, and the object features can also be composed of other features. The user attribute features can also be composed in other ways. For example, the user age feature can be divided into 0 to 10 years old, 11 to 15 years old, 16 to 25 years old, 26 to 40 years old, and over 41 years old, etc. That is, the composition data of the user features and object features given above is only a possible case for illustrative purposes and does not limit the combination methods of other sub-features and the division methods of each sub-feature.

[0151] Further, the computer device can perform vector conversion on at least two sub-features in the i-th concatenated feature 501 based on the embedding layer to obtain conversion vectors respectively corresponding to the at least two sub-features. As Figure 5As shown in the figure, the computer device can perform vector transformation on the sub-feature 5011 based on the embedding layer to obtain the transformation vector 5021 corresponding to the sub-feature 5011; perform vector transformation on the sub-feature 5012 to obtain the transformation vector 5022 corresponding to the sub-feature 5012;...; perform vector transformation on the sub-feature 5013 to obtain the transformation vector 5023 corresponding to the sub-feature 5013. The computer device performs feature fusion on at least two sub-features in the i-th concatenated feature 501 based on the factorization layer to obtain the i-th first fusion feature 5031. Specifically, feature fusion is performed on the sub-feature 5011, the sub-feature 5012,..., and the sub-feature 5013 to obtain the i-th first fusion feature 5031. The computer device can perform feature inner product processing on the transformation vectors corresponding to at least two sub-features 501 respectively based on the factorization layer to obtain the i-th second fusion feature 5032. Specifically, feature inner product processing is performed on the transformation vector 5021, the transformation vector 5022,..., and the transformation vector 5023 to obtain the i-th second fusion feature 5032. The computer device can perform activation processing on the transformation vectors corresponding to at least two sub-features 501 respectively based on the hidden layer to obtain the i-th third fusion feature 5033. Specifically, activation processing is performed on the transformation vector 5021, the transformation vector 5022,..., and the transformation vector 5023 to obtain the i-th third fusion feature 5033. The computer device performs association prediction on the i-th first fusion feature 5031, the i-th second fusion feature 5032, and the i-th third fusion feature 5033 based on the output layer to obtain the predicted association degree between the i-th candidate object and the target user. Among them, ⊕ is used to represent the feature fusion operation (Addition), is used to represent the feature inner product processing (Inner Product), is used to represent the activation processing (Activation Function), is used to represent the activation or logistic regression processing, such as the sigmoid function, etc.

[0152] Among them, the computer device can process the transformation vectors corresponding to at least two sub-features in the i-th concatenated feature respectively, perform association prediction on each transformation vector and between each transformation vector to obtain the i-th second fusion feature 5032, that is, it is equivalent to processing each sub-feature and performing association processing between each pair of sub-features to consider the relationship between the object feature of the i-th candidate object and the target user feature of the target user, and perform association prediction on the i-th candidate object and the target user, so as to improve the accuracy of the association degree prediction of the data recommendation model for each candidate object and the target user. Among them, the generation process of the i-th second fusion feature 5032 can be seen in formula ③ as follows:

[0153]

[0154] Among them, g is used to represent the number of features of at least two sub-features included in the i-th splicing feature 501, or it can be considered as the number of vectors of the transformation vectors respectively corresponding to the at least two sub-features. Among them, x l is used to represent the l-th transformation vector among the transformation vectors respectively corresponding to the at least two sub-features, x o is used to represent the o-th transformation vector among the transformation vectors respectively corresponding to the at least two sub-features. Among them, ω0, ω l and ω lo are model parameters in the data recommendation model. Among them, ω0 can be used to represent the initial weight in the data recommendation model, ω l can be used to represent the weight for the transformation vector, ω lo can be used to represent the combined weight between every two of the transformation vectors respectively corresponding to the at least two sub-features.

[0155] Furthermore, based on the determination process of the predicted association degree between the i-th candidate object and the target user, determine the predicted association degrees between at least two candidate objects and the target user respectively.

[0156] Furthermore, the computer device can obtain a recommended quantity threshold, and based on the predicted association degree, perform an association ranking on at least two candidate objects, and obtain N candidate objects from the at least two candidate objects after the association ranking as the target recommended objects for the target user; N is a positive integer, and N is greater than or equal to the recommended quantity threshold. Among them, N generally can take the recommended quantity threshold. The computer device can output the target recommended objects for the target user when the target user logs in; optionally, when N takes a value greater than the recommended quantity threshold, N can also be less than or equal to the recommended upper limit value. The computer device can output the candidate objects corresponding to the recommended quantity threshold in the target recommended objects in sequence when the target user logs in. If it is detected that there is an abnormal output of a candidate object, then obtain candidate objects from the target recommended objects in sequence for output. For example, the recommended quantity threshold is 5 and N is 7, that is, the target recommended objects include 7 candidate objects. The computer device responds to the login operation for the target user and outputs the first 5 candidate objects in the target recommended objects. If it is detected that there is an abnormal output of a candidate object among the first 5 candidate objects, then output the 6th candidate object and the 7th candidate object in the target recommended objects to the target user in sequence until the target user can obtain 5 normal candidate objects, thereby improving the fault tolerance of data recommendation.

[0157] Among them, the target user and the target recommended object (which can also be referred to as the object recommendation list) are stored in the offline layer. The computer device sends the target user and the target recommended object of the target user to the online layer based on the offline layer, and stores the target user and the target recommended object in an associated manner in the user recommendation library in the online layer. In the online layer, the computer device responds to the login operation for the target user, obtains the target recommended object associated with the target user from the user recommendation library based on the associated storage relationship, and outputs the target recommended object. Specifically, the target recommended object is sent to the target user terminal where the target user is located, so that the target user terminal displays the target recommended object.

[0158] Further, in the online layer, the computer device responds to the login operation for the user to be processed, and searches for the recommended object to be output associated with the user to be processed from the user recommendation library based on the associated storage relationship. If there is no recommended object to be output associated with the user to be processed in the user recommendation library, the default recommended object is obtained, and the default recommended object is output for the user to be processed; optionally, if there is no recommended object to be output associated with the user to be processed in the user recommendation library, the computer device can also perform object recall on at least two candidate objects to obtain the recalled object, perform association prediction on the recalled object and the object to be processed to obtain the recommended object to be output for the object to be processed, and output the recommended object to be output for the user to be processed. Among them, when the computer device performs object recall on at least two candidate objects to obtain the recalled object, it can obtain the characteristics of the user to be processed of the user to be processed, and perform object recall on at least two candidate objects based on the characteristics of the user to be processed to obtain the recalled object. For example, the gender and age of the user to be processed are obtained, the user cluster corresponding to the gender and age of the user to be processed is obtained, and object recall is performed on at least two candidate objects based on the associated objects of the user cluster to obtain the recalled object corresponding to the user to be processed.

[0159] In an embodiment of the present application, a computer device may obtain historical login status information of a target user, and predict the predicted login status of the target user within a target time period according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online status, search for target user features associated with the target user from a user feature library, and obtain object features corresponding to at least two candidate objects respectively; perform an association prediction on the object features corresponding to the at least two candidate objects respectively and the target user features to obtain the predicted association degrees of the at least two candidate objects with the target user respectively, and based on the predicted association degrees, determine a target recommendation object of the target user from the at least two candidate objects; the target recommendation object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status. By predicting the login status of the target user, the predicted login status of the target user is obtained. When the predicted login status belongs to the online status, the target recommendation object of the target user can be determined, so that the target recommendation object can be determined in advance, and when the target user logs in, the target recommendation object predicted in advance can be directly obtained, thereby improving the efficiency of data recommendation. Further, in the present application, the association degrees of at least two candidate objects with the target user are predicted respectively, and full-object prediction scoring (i.e., predicted association degree) is achieved, improving the accuracy of data recommendation.

[0160] Further, reference may be made to Figure 6 , Figure 6 which is a schematic diagram of the training process of a data recommendation model provided by an embodiment of the present application. As Figure 6 shown, the training process of the data recommendation model may include the following steps:

[0161] Step S601, obtain at least two training user samples, and obtain sample user features respectively associated with the at least two training user samples from a user feature library.

[0162] In an embodiment of the present application, a computer device may, within a training period, obtain at least two candidate users associated with an application program, where the application program is used to output a target recommendation object for a target user. The computer device may obtain the candidate login statuses corresponding to the at least two candidate users during the training period, and determine the candidate users with an online status in the candidate login statuses as training user samples. The user feature library includes at least two statistical users and the user features of each statistical user. The at least two statistical users include statistical user j; j is a positive integer, and j is less than or equal to the number of users in the at least two statistical users. The computer device may obtain the training sample hash values corresponding to the at least two training user samples respectively, and generate a sample matching bit array according to the training sample hash values corresponding to the at least two training user samples respectively; generate the user hash value of statistical user j, and obtain the user status value of statistical user j from the sample matching bit array according to the user hash value of statistical user j; if there is at least one missing status value in the user status value of statistical user j, it is determined that statistical user j is not associated with the at least two training user samples, and if all the user status values of statistical user j are valid status values, it is determined that statistical user j is associated with the at least two training user samples; the user features corresponding to the statistical users associated with the at least two training user samples in the user feature library are determined as the sample user features associated with the at least two training user samples. Specifically, the computer device obtains the sample user features associated with the at least two training user samples from the user features corresponding to the statistical users associated with the at least two training user samples in the user feature library. Among them, the computer device may screen the at least two statistical users based on the sample matching bit array, which can remove a large amount of data, and then find the at least two training user samples from the screened statistical users, which can reduce the amount of data to be processed. Since the storage space occupied by the sample matching bit array and the resources or time consumed in the matching process are relatively small, the acquisition efficiency of the sample user features associated with the at least two training user samples can be improved, and thus the training efficiency of the data recommendation model can be improved.

[0163] Specifically, the sample matching bit array is generated based on at least one hash function. The sample matching bit array can be a binary array with valid state values and missing state values respectively. Assume the length of the sample matching bit array is d, and the number of at least one hash function is e. Both d and e are positive integers. Among them, an initial sample matching bit array is generated, and the d state values in the initial sample matching bit array are all set to missing state values. Among them, when generating the sample matching bit array according to at least two training user samples, each training user sample is hashed based on e hash functions to obtain e state hash values corresponding to each training user sample respectively. The e state hash values respectively correspond to e state positions in the initial sample matching bit array, and the state values at the e state positions are updated to valid state values to generate the sample matching bit array. Among them, the state position can be the state hash value obtained according to the hash function, or can be obtained by converting the state hash value, such as conversion methods like taking the remainder. Among them, when matching at least two statistical users, the computer device can obtain e user hash values of the statistical user j according to e hash functions, obtain e statistical state positions corresponding to the e user hash values in the sample matching bit array, and obtain the user state value corresponding to the statistical user j according to the e statistical state positions. If there is a missing state value in the user state value of the statistical user j, it is determined that the statistical user j is not associated with at least two training user samples; if all the user state values of the statistical user j are valid state values, it is determined that the statistical user j is associated with at least two training user samples.

[0164] Specifically, reference can be made to Figure 7 , Figure 7 which is a schematic diagram of the generation and filtering scenario of a sample matching bit array provided by an embodiment of the present application. As Figure 7 shown, assume that there are 3 elements {a, b, c} in at least two training user samples, and e is 3, the valid state value is 1, and the missing state value is 0. When all the state values in the initial sample matching bit array are missing state values 0, 3 state hash values of element a are obtained through 3 hash functions. The 3 state hash values respectively correspond to bit 1, bit 5, and bit 13 in the initial sample matching bit array, and the state values at bit 1, bit 5, and bit 13 in the initial sample matching bit array are set to 1; 3 state hash values of element b are obtained through 3 hash functions. The 3 state hash values respectively correspond to bit 4, bit 11, and bit 16 in the initial sample matching bit array, and the state values at bit 4, bit 11, and bit 16 in the initial sample matching bit array are set to 1; 3 state hash values of element c are obtained through 3 hash functions. The 3 state hash values respectively correspond to bit 3, bit 5, and bit 11 in the initial sample matching bit array, and the state values at bit 3, bit 5, and bit 11 in the initial sample matching bit array are set to 1, obtaining Figure 7The sample matching bit array shown in Figure 7 the sample matching bit array shown in. Optionally, the process can also be implemented based on other programming languages, which is not limited here.

[0165] When element f is obtained, three state hash values of element f are obtained through the above three hash functions. These three state hash values respectively correspond to bits 3, 4, and 5 in the sample matching bit array. If the state values at bits 3, 4, and 5 in the sample matching bit array are all valid state value 1, it is determined that element f exists in the sample matching bit array, but in fact, element f does not exist in this mapping relationship. When element w is obtained, three state hash values of element w are obtained through the above three hash functions. These three state hash values respectively correspond to bits 4, 13, and 15 in the sample matching bit array. If the state values at bits 4 and 13 in the sample matching bit array are both valid state value 1, and the state value at bit 15 is the missing state value 0, it is determined that element w does not exist in the sample matching bit array. It can be seen that when three state values are obtained through the above three hash functions, if there is a missing state value 0 among these three state values, it can be determined that the element corresponding to these three state values must not be in the sample matching bit array. When the three state values are all valid state value 1, the element corresponding to these three state values is not necessarily in the sample matching bit array. Therefore, there may be a certain misjudgment rate. Thus, when it is determined that the statistical user j exists in the sample matching bit array, the statistical user j may not necessarily exist in at least two training user samples. When it is determined that the statistical user j does not exist in the sample matching bit array, the statistical user j must not exist in at least two training user samples. Therefore, through this process, when obtaining the sample user features corresponding to at least two training user samples respectively, the amount of data to be processed can be reduced, and the efficiency of model training can be improved.

[0166] Optionally, when the computer device obtains statistical users associated with at least two training user samples among at least two statistical users, it can be implemented based on a distributed computing engine such as Spark, etc., which is not limited here. Among them, Spark is a fast computing engine for large-scale data processing and can be used to build large-scale and low-latency data analysis applications.

[0167] Step S602: Obtain the sample object features associated with each training user sample, and perform model training based on the sample user features and sample object features respectively corresponding to at least two training user samples to generate a data recommendation model.

[0168] In the embodiment of the present application, the computer device can count the historical associated objects generated by the training user samples in the online state, and determine the historical object features of the historical associated objects as the sample object features associated with the training user samples. Further, the at least two training user samples include training user sample k; k is a positive integer, and k is less than or equal to the sample quantity of the at least two training user samples. The computer device can predict the sample user features and sample object features of the training user sample k based on the initial data recommendation model to obtain the sample prediction correlation degree between the training user sample k and the sample object features of the training user sample k; train the initial data recommendation model based on the sample prediction correlation degree and the model loss function to generate a data recommendation model. Among them, the model loss function (Logit Loss Function) refers to a function such as the cross-entropy loss function, etc., which is not limited here.

[0169] In the embodiment of the present application, the computer device can obtain at least two training user samples, perform model training based on the sample object features and sample user features associated with the training user samples to generate a data recommendation model. The data recommendation model is used to predict the correlation degree between at least two candidate objects and the target user respectively. The data recommendation model trains the relationship between the sample user features and the sample object features, so that the relationship between the user features and the object features can be represented in the data recommendation model, thereby improving the prediction accuracy of the data recommendation model for the predicted correlation degree.

[0170] Further, please refer to Figure 8 , Figure 8 which is a schematic diagram of a data recommendation architecture provided by an embodiment of the present application. As Figure 8As shown, the computer device is divided into an online layer and an offline layer. Among them, the offline layer has a large storage space and can perform large-scale data processing. The computer device can obtain at least two candidate users associated with the application program during the training period in the offline layer, obtain the corresponding candidate login statuses of the at least two candidate users during the training period, and determine the candidate users with the online status in the candidate login statuses as training user samples. Obtain the sample user features of the training user samples from the user feature library. Among them, this process can be to traverse the user feature library to obtain the sample user features of the training user samples, or to generate a sample matching bit array according to the training sample hash values respectively corresponding to the at least two training user samples, obtain the user status values of each statistical user in the user feature library in the sample matching bit array, and determine the statistical users with the user status values all being valid status values as the statistical users associated with the at least two training user samples. Based on the statistical users associated with the at least two training user samples, determine the sample user features respectively corresponding to the at least two training user samples. The computer device can obtain the sample object features associated with the training user samples from the object feature library, and train a model based on the sample user features and sample object features respectively corresponding to the at least two training user samples to generate a data recommendation model.

[0171] Furthermore, the computer device can obtain a target user, obtain the historical login status information of the target user, and predict the predicted login status of the target user within the target time period based on the historical login status information. Among them, the number of the target users is one or at least two. Specifically, the computer device can obtain the historical login status generated by the target user during the status prediction period, determine the historical login status information of the target user based on the historical login status generated during the status prediction period, and obtain at least two default distribution states; it can determine the state change sequence according to the historical login status information, use the at least two default distribution states as state nodes respectively, determine the edges between the at least two state nodes based on the state change sequence to generate a state chain; determine the state transition matrix based on the state chain. Based on the characteristics of the prediction probability, the computer device can determine the predicted login status of the target user within the target time period according to the state transition matrix.

[0172] The computer device can perform batch processing on target users. It can record the target users whose predicted login status belongs to the online status as online target users, call a data recommendation model to perform association prediction on at least two candidate objects and the online target users, obtain the predicted association degrees between each of the at least two candidate objects and the online target users respectively, determine the target recommended objects of the online target users from the at least two candidate objects based on the predicted association degrees between each of the at least two candidate objects and the online users respectively, perform associated storage on the online target users and the target recommended objects to obtain an object recommendation list, and send the object recommendation list to the online layer. In the online layer, the computer device stores the object recommendation list in the user recommendation library. When the computer device detects a user login, assuming that the user is a target user whose predicted login status is the online status, the computer device can obtain the target recommended objects associated with the user from the user recommendation library, re-rank based on the target recommended objects, and output the re-ranked target recommended objects for the user. Optionally, the computer device can store data such as the user feature library and the object feature library based on big data technology, and can use an offline database to store data, such as Hadoop, etc. Among them, Hadoop is a distributed system infrastructure that can achieve high-speed operation and storage.

[0173] The computer device realizes by placing each process of data recommendation in the offline layer, such as the generation process of the data recommendation model, the prediction process of the predicted login status of the target user, and the prediction process of the target recommended objects of the target users whose predicted login status belongs to the online status, etc., so that the computer device does not need to store the data recommendation model, the user feature library, the object feature library, etc. in the online layer, reducing the data synchronization pressure between the offline layer and the online layer and alleviating the data processing work pressure of the online layer. And because the offline layer has a large storage space and can perform large-scale data processing, therefore, each process of this data recommendation will not cause work pressure on the offline layer, thus improving the efficiency of data recommendation. Moreover, the computer device will obtain the predicted association degrees between each of the at least two candidate objects and the target user respectively to determine the target recommended objects among the at least two candidate objects, enabling full-object prediction of the target recommended objects associated with the target user, reducing the risk of accuracy reduction caused by object recall, and improving the accuracy of data recommendation. Additionally, the computer device can predict the predicted login status of the target user based on the historical login status information of the target user, determine the users who may be online during the target time period based on the predicted login status, perform early prediction on these users who may be online during the target time period, store the target recommended objects of these users who may be online during the target time period, so that when these users log in, the computer device can directly obtain the target recommended objects associated with these users and output the associated target recommended objects for these users, thereby improving the real-time performance of object output and the efficiency of data recommendation.

[0174] Among them, when the computer device detects the login operation of the user to be processed, if there is no to-be-output recommended object associated with the user to be processed in the user recommendation library, the default recommended object can be obtained and the default recommended object can be output for the user to be processed; or, the user to be processed can be predicted based on the data recommendation model to obtain the to-be-output recommended object of the user to be processed. Among them, taking predicting the user to be processed based on the data recommendation model to obtain the to-be-output recommended object of the user to be processed as an example, reference can be made to Figure 9 , Figure 9 which is another schematic diagram of the data recommendation architecture provided by the embodiments of the present application. As Figure 9 shown, the computer device can obtain training user samples, obtain the sample user features of the training user samples from the user feature library, obtain the sample object features of the training user samples from the object feature library, and train a model based on the sample user features and sample object features of the training user samples to obtain a data recommendation model.

[0175] When the computer device detects the login operation of the user to be processed, in response to the login operation of the user to be processed, search for the to-be-output recommended object associated with the user to be processed in the user recommendation library. Assuming that there is no to-be-output recommended object associated with the user to be processed in the user recommendation library, the computer device can obtain the data recommendation model, the object feature library, and the user feature library from the offline layer, and store the data recommendation model, the object feature library, and the user feature library in the online layer. In the online layer, the computer device obtains the to-be-processed user features of the user to be processed from the user feature library, performs object recall on at least two candidate objects in the object feature library to obtain the recalled objects, obtains the recalled object features of the recalled objects, and performs an association prediction on the recalled object features and the to-be-processed user features based on the data recommendation model to determine the predicted association degree between the recalled objects and the user to be processed. Among them, the number of the recalled objects is not unique, and the to-be-output recommended object of the user to be processed is determined based on the predicted association degree between the recalled objects and the user to be processed. Rearrange based on the to-be-output recommended object, and output the rearranged to-be-output recommended object for the user to be processed.

[0176] Further, please refer to Figure 10 , Figure 10 which is a schematic diagram of a data recommendation device provided by the embodiments of the present application. The data recommendation device can be a computer program (including program codes, etc.) running in a computer device. For example, the data recommendation device can be an application software; the device can be used to execute the corresponding steps in the method provided by the embodiments of the present application. As Figure 10 shown, the data recommendation device 1000 can be used to Figure 3The computer device in the corresponding embodiment. Specifically, the device may include: a status prediction module 11, a feature acquisition module 12, and a recommendation prediction module 13.

[0177] The status prediction module 11 is configured to obtain the historical login status information of the target user, and predict the predicted login status of the target user within the target time period according to the historical login status information.

[0178] The feature acquisition module 12 is configured to, if the predicted login status of the target user within the target time period belongs to the online status, search for the target user features associated with the target user from the user feature library, and obtain the object features corresponding to at least two candidate objects.

[0179] The recommendation prediction module 13 is configured to perform an association prediction on the object features corresponding to at least two candidate objects and the target user features, obtain the predicted association degrees between at least two candidate objects and the target user respectively, and based on the predicted association degrees, determine the target recommendation object of the target user from at least two candidate objects; the target recommendation object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status.

[0180] Among them, the status prediction module 11 includes:

[0181] The distribution acquisition unit 111 is configured to obtain the historical login status information of the target user and obtain at least two default distribution states.

[0182] The matrix acquisition unit 112 is configured to obtain a state transition matrix according to the historical login status information and at least two default distribution states.

[0183] The status prediction unit 113 is configured to predict the predicted login status of the target user within the target time period according to the state transition matrix.

[0184] Among them, the matrix acquisition unit 112 includes:

[0185] The graph generation sub-unit 1121 is configured to determine a state change sequence according to the historical login status information, use at least two default distribution states as state nodes respectively, determine the edges between at least two state nodes based on the state change sequence, and generate a state chain.

[0186] The matrix determination sub-unit 1122 is configured to determine a state transition matrix based on the state chain.

[0187] Among them, the matrix determination sub-unit 1122 is specifically configured to:

[0188] Determine the matrix dimension feature based on at least two state nodes in the state chain, determine the matrix parameter according to the edges between at least two state nodes in the state chain, and form an initial state transition matrix with the matrix dimension feature and the matrix parameter;

[0189] Perform a transfer process on the initial state transition matrix to obtain a state transition matrix.

[0190] Among them, in terms of performing an association prediction on the object features corresponding to at least two candidate objects and the target user feature respectively to obtain the prediction association degrees between at least two candidate objects and the target user respectively, the recommendation prediction module 13 includes:

[0191] A feature splicing unit 131, configured to splice the object feature of the i-th candidate object among at least two candidate objects and the target user feature to obtain the i-th spliced feature; i is a positive integer, and i is less than or equal to the candidate quantity of at least two candidate objects;

[0192] An association prediction unit 132, configured to perform an association prediction on the i-th spliced feature based on the data recommendation model to obtain the prediction association degree between the i-th candidate object and the target user, until the prediction association degrees between at least two candidate objects and the target user respectively are obtained.

[0193] Among them, the association prediction unit 132 includes:

[0194] A vector conversion sub-unit 1321, configured to perform vector conversion on at least two sub-features in the i-th spliced feature based on the data recommendation model to obtain conversion vectors corresponding to at least two sub-features respectively;

[0195] A feature fusion sub-unit 1322, configured to perform feature fusion on at least two sub-features in the i-th spliced feature to obtain the i-th first fusion feature, perform feature inner product processing on the conversion vectors corresponding to at least two sub-features respectively to obtain the i-th second fusion feature, and perform activation processing on the conversion vectors corresponding to at least two sub-features respectively to obtain the i-th third fusion feature;

[0196] A feature prediction sub-unit 1323, configured to perform an association prediction on the i-th first fusion feature, the i-th second fusion feature, and the i-th third fusion feature to obtain the prediction association degree between the i-th candidate object and the target user.

[0197] Among them, in terms of determining the target recommendation object of the target user from at least two candidate objects based on the prediction association degree, the recommendation prediction module 13 is specifically configured to:

[0198] Obtain a recommended quantity threshold, perform an association ranking on at least two candidate objects based on the predicted association degree, and obtain N candidate objects from the at least two candidate objects after the association ranking as the target recommended objects for the target user; N is a positive integer, and N is greater than or equal to the recommended quantity threshold.

[0199] Among them, the target user and the target recommended objects are stored in the offline layer; the apparatus 1000 further includes:

[0200] A recommended storage module 14, configured to send the target user and the target recommended objects of the target user to the online layer based on the offline layer, and perform an associated storage of the target user and the target recommended objects in the user recommendation library in the online layer;

[0201] A recommended output module 15, configured to, in the online layer, in response to a login operation for the target user, obtain the target recommended objects associated with the target user from the user recommendation library based on the associated storage relationship, and output the target recommended objects.

[0202] Among them, the apparatus 1000 further includes:

[0203] An object search module 16, configured to, in the online layer, in response to a login operation for the to-be-processed user, search for the to-be-output recommended objects associated with the to-be-processed user from the user recommendation library based on the associated storage relationship;

[0204] A default output module 17, configured to, if there are no to-be-output recommended objects associated with the to-be-processed user in the user recommendation library, obtain default recommended objects and output the default recommended objects for the to-be-processed user.

[0205] Among them, the apparatus 1000 further includes:

[0206] A sample acquisition module 18, configured to acquire at least two training user samples, and acquire the sample user features respectively associated with the at least two training user samples from the user feature library;

[0207] A model training module 19, configured to acquire the sample object features associated with each training user sample, and perform model training based on the sample user features and the sample object features respectively corresponding to the at least two training user samples to generate a data recommendation model.

[0208] Among them, in terms of acquiring at least two training user samples, the sample acquisition module 18 includes:

[0209] A candidate acquisition unit 181, configured to acquire at least two candidate users associated with the application program within a training period; the application program is used to output the target recommended objects of the target user;

[0210] A sample selection unit 182 is configured to obtain candidate login states corresponding to at least two candidate users during a training period, and determine candidate users whose candidate login states include an online state as training user samples;

[0211] In terms of obtaining sample object features associated with each training user sample, the model training module 19 includes:

[0212] An object acquisition unit 191 is configured to count historical associated objects generated by a training user sample in an online state, and determine historical object features of the historical associated objects as sample object features associated with the training user sample.

[0213] Wherein, the user feature library includes at least two statistical users and user features of each statistical user; the at least two statistical users include a statistical user j; j is a positive integer, and j is less than or equal to the number of users of the at least two statistical users;

[0214] In terms of obtaining sample user features respectively associated with at least two training user samples from the user feature library, the sample acquisition module 18 includes:

[0215] An array generation unit 183 is configured to obtain training sample hash values respectively corresponding to at least two training user samples, and generate a sample matching bit array according to the training sample hash values respectively corresponding to the at least two training user samples;

[0216] A hash matching unit 184 is configured to generate a user hash value of the statistical user j, and obtain a user status value of the statistical user j from the sample matching bit array according to the user hash value of the statistical user j;

[0217] An association determination unit 185 is configured to determine that the statistical user j is not associated with the at least two training user samples if there is at least one missing status value in the user status value of the statistical user j, and determine that the statistical user j is associated with the at least two training user samples if the user status values of the statistical user j are all valid status values;

[0218] A feature determination unit 186 is configured to determine user features corresponding to statistical users associated with at least two training user samples in the user feature library as sample user features associated with the at least two training user samples.

[0219] Wherein, the at least two training user samples include a training user sample k; k is a positive integer, and k is less than or equal to the number of samples of the at least two training user samples;

[0220] In terms of performing model training based on sample user features and sample object features respectively corresponding to at least two training user samples to generate a data recommendation model, the model training module 19 includes:

[0221] A sample prediction unit 192 is configured to predict the sample user features and sample object features of the training user sample k based on the initial data recommendation model, and obtain the sample prediction correlation degree between the training user sample k and the sample object features of the training user sample k;

[0222] A model training unit 193 is configured to train the initial data recommendation model based on the sample prediction correlation degree and the model loss function to generate a data recommendation model.

[0223] An embodiment of the present application provides a data recommendation device. The device can obtain the historical login status information of a target user, and predict the predicted login status of the target user within a target time period according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online status, then search for the target user features associated with the target user from a user feature library, and obtain the object features corresponding to at least two candidate objects respectively; perform an association prediction on the object features corresponding to the at least two candidate objects and the target user features to obtain the prediction correlation degrees between the at least two candidate objects and the target user respectively, and based on the prediction correlation degrees, determine the target recommendation object of the target user from the at least two candidate objects; the target recommendation object is used as the recommended content of the target user when it is detected that the actual login status of the target user within the target time period is the online status. By predicting the login status of the target user, the predicted login status of the target user is obtained. When the predicted login status belongs to the online status, the target recommendation object of the target user can be determined, so that the target recommendation object can be determined in advance, and when the target user logs in, the target recommendation object predicted in advance can be directly obtained, thereby improving the efficiency of data recommendation. Further, in the present application, the association degree prediction is performed on each of the at least two candidate objects and the target user respectively, realizing the full-object prediction scoring (i.e., the prediction correlation degree), and improving the accuracy of data recommendation.

[0224] See Figure 11 , Figure 11 is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 11 shown, the computer device in the embodiment of the present application may include: one or more processors 1101, a memory 1102, and an input / output interface 1103. The processor 1101, the memory 1102, and the input / output interface 1103 are connected through a bus 1104. The memory 1102 is used to store a computer program, and the computer program includes program instructions. The input / output interface 1103 is used to receive data and output data, such as for data interaction with a user device, or for data interaction between an online layer and an offline layer in the computer device; the processor 1101 is used to call the computer program stored in the memory 1102 to enable the computer device to execute the data recommendation method in the present application.

[0225] Among them, the processor 1101 can perform the following operations:

[0226] Obtain the historical login status information of the target user, and predict the predicted login status of the target user within the target time period according to the historical login status information;

[0227] If the predicted login status of the target user within the target time period belongs to the online status, search for the target user features associated with the target user in the user feature library, and obtain the object features corresponding to at least two candidate objects;

[0228] Perform an association prediction on the object features corresponding to at least two candidate objects and the target user features to obtain the predicted association degrees between at least two candidate objects and the target user respectively. Based on the predicted association degrees, determine the target recommended object of the target user from at least two candidate objects; the target recommended object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status.

[0229] Among them, obtaining the historical login status information of the target user and predicting the predicted login status of the target user within the target time period according to the historical login status information includes:

[0230] Obtain the historical login status information of the target user and obtain at least two default distribution states;

[0231] According to the historical login status information and at least two default distribution states, obtain a state transition matrix;

[0232] Predict the predicted login status of the target user within the target time period according to the state transition matrix.

[0233] Among them, according to the historical login status information and at least two default distribution states, obtaining a state transition matrix includes:

[0234] Determine the state change sequence according to the historical login status information, use at least two default distribution states as state nodes respectively, determine the edges between at least two state nodes based on the state change sequence, and generate a state chain;

[0235] Determine the state transition matrix based on the state chain.

[0236] In some possible embodiments, the processor 1101 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0237] The memory 1102 may include a read-only memory and a random access memory, and provide instructions and data to the processor 1101 and the input / output interface 1103. A part of the memory 1102 may also include a non-volatile random access memory. For example, the memory 1102 may also store information about the device type.

[0238] In a specific implementation, the computer device may execute, through each of its built-in functional modules, the implementation manners provided in each step of the Figure 3 For details, reference may be made to the implementation manners provided in each step of the Figure 3 For details, reference may be made to the implementation manners provided in each step of the

[0239] By providing a computer device according to an embodiment of the present application, including: a processor, an input / output interface, and a memory, the computer program in the memory is obtained through the processor and executed. Figure 3For each step of the method shown, perform a data recommendation operation. In the embodiments of the present application, the historical login status information of the target user is obtained, and the predicted login status of the target user within the target time period is predicted according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online status, the target user features associated with the target user are searched from the user feature library, and the object features corresponding to at least two candidate objects are obtained; the object features corresponding to at least two candidate objects are associated and predicted with the target user features to obtain the predicted association degrees between at least two candidate objects and the target user respectively, and based on the predicted association degrees, the target recommended object of the target user is determined from at least two candidate objects; the target recommended object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status. By predicting the login status of the target user, the predicted login status of the target user is obtained. When the predicted login status belongs to the online status, the target recommended object of the target user can be determined, so that the target recommended object can be determined in advance, and when the target user logs in, the target recommended object predicted in advance can be directly obtained, thereby improving the efficiency of data recommendation. Further, in the present application, the association degrees between at least two candidate objects and the target user are predicted respectively, and the full-object prediction scoring (i.e., the predicted association degree) is realized, which improves the accuracy of data recommendation.

[0240] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program, and the computer program is suitable for being loaded and executed by the processor Figure 3 in the data recommendation method provided by each step, and specifically, reference can be made to the Figure 3 implementation manners provided by each step, which will not be elaborated here. In addition, the description of the beneficial effects of adopting the same method will not be elaborated either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application. As an example, the computer program can be deployed to be executed on one computer device, or on multiple computer devices located at one place, or on multiple computer devices distributed at multiple places and interconnected through a communication network.

[0241] The computer-readable storage medium may be the data recommendation device provided in any of the foregoing embodiments or the internal storage unit of the computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the computer-readable storage medium may also include both the internal storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store the data that has been output or will be output.

[0242] The embodiments of the present application also provide a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 3 the methods provided in various alternative manners in [description of the methods], realizes obtaining the historical login status information of the target user, predicting the predicted login status of the target user within the target time period according to the historical login status information; if the predicted login status of the target user within the target time period belongs to the online status, searching for the target user characteristics associated with the target user from the user feature library, and obtaining the object characteristics corresponding to at least two candidate objects respectively; performing an association prediction on the object characteristics corresponding to at least two candidate objects respectively and the target user characteristics to obtain the predicted association degrees between at least two candidate objects and the target user respectively, and based on the predicted association degrees, determining the target recommended object of the target user from at least two candidate objects; the target recommended object is used as the recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status. By predicting the login status of the target user, the predicted login status of the target user is obtained. When the predicted login status belongs to the online status, the target recommended object of the target user can be determined, so that the target recommended object can be determined in advance, and when the target user logs in, the target recommended object that has been predicted in advance can be directly obtained, thereby improving the efficiency of data recommendation. Further, in the present application, the association degrees between at least two candidate objects and the target user are predicted respectively, realizing full-object prediction scoring (i.e., predicted association degrees), and improving the accuracy of data recommendation.

[0243] In the description, claims, and drawings of the embodiments of this application, the terms "first", "second", etc. are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or modules, but may optionally further include steps or modules not listed, or may optionally further include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0244] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in this description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0245] The methods and related apparatuses provided in the embodiments of this application are described with reference to the method flowcharts and / or structural schematic diagrams provided in the embodiments of this application. Specifically, each process and / or block of the method flowchart and / or structural schematic diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data recommendation devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data recommendation devices generate an apparatus for implementing the functions specified in Figure 1 one process or multiple processes and / or structural schematic Figure 1 one block or multiple blocks. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data recommendation device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction apparatus, and the instruction apparatus implements the functions specified in Figure 1 one process or multiple processes and / or structural schematic Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data recommendation device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or structural schematic one block or multiple blocks.

[0246] The above disclosure is only for the preferred embodiments of the present application. Of course, the scope of rights of the present application cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A data recommendation method, characterized in that, The method includes: Obtaining historical login status information of a target user, and predicting a predicted login status of the target user within a target time period according to the historical login status information; If the predicted login status of the target user within the target time period belongs to the online status, searching for target user features associated with the target user from a user feature library, and obtaining object features corresponding to at least two candidate objects respectively; Performing an association prediction on the object features corresponding to the at least two candidate objects and the target user features, obtaining a predicted association degree between each of the at least two candidate objects and the target user, and determining a target recommended object of the target user from the at least two candidate objects based on the predicted association degree; the target recommended object is used as recommended content for the target user when it is detected that the actual login status of the target user within the target time period is the online status; If the predicted login status of the target user within the target time period belongs to the offline status, no processing is performed on the target user, or a default recommended object is determined as the target recommended object of the target user.

2. The method according to claim 1, characterized in that The obtaining historical login status information of a target user and predicting a predicted login status of the target user within a target time period according to the historical login status information includes: Obtaining historical login status information of a target user, and obtaining at least two default distribution states; Obtaining a state transition matrix according to the historical login status information and the at least two default distribution states; Predicting a predicted login status of the target user within the target time period according to the state transition matrix.

3. The method according to claim 2, wherein The obtaining a state transition matrix according to the historical login status information and the at least two default distribution states includes: Determining a state change sequence according to the historical login status information, using the at least two default distribution states as state nodes respectively, determining edges between the at least two state nodes based on the state change sequence, and generating a state chain; Determining a state transition matrix based on the state chain.

4. The method according to claim 3, characterized in that, The determining a state transition matrix based on the state chain includes: Determining matrix dimension features based on at least two state nodes in the state chain, determining matrix parameters according to edges between the at least two state nodes in the state chain, and forming an initial state transition matrix with the matrix dimension features and the matrix parameters; Performing a transition process on the initial state transition matrix to obtain a state transition matrix.

5. The method according to claim 1, wherein The performing an association prediction on the object features corresponding to the at least two candidate objects and the target user features, and obtaining a predicted association degree between each of the at least two candidate objects and the target user includes: Performing feature splicing on the object feature of the i-th candidate object among the at least two candidate objects and the target user features to obtain an i-th spliced feature; i is a positive integer, and i is less than or equal to the number of candidates of the at least two candidate objects; Based on the data recommendation model, perform association prediction on the i-th spliced feature to obtain the predicted association degree between the i-th candidate object and the target user, until the predicted association degrees between the at least two candidate objects and the target user are obtained respectively.

6. The method according to claim 5, wherein The performing association prediction on the i-th spliced feature based on the data recommendation model to obtain the predicted association degree between the i-th candidate object and the target user includes: Based on the data recommendation model, perform vector transformation on at least two sub-features in the i-th spliced feature to obtain transformation vectors corresponding to the at least two sub-features respectively; Perform feature fusion on at least two sub-features in the i-th spliced feature to obtain the i-th first fusion feature, perform feature inner product processing on the transformation vectors corresponding to the at least two sub-features respectively to obtain the i-th second fusion feature, and perform activation processing on the transformation vectors corresponding to the at least two sub-features respectively to obtain the i-th third fusion feature; Perform association prediction on the i-th first fusion feature, the i-th second fusion feature, and the i-th third fusion feature to obtain the predicted association degree between the i-th candidate object and the target user.

7. The method according to claim 1, characterized in that, The determining the target recommendation object of the target user from the at least two candidate objects based on the predicted association degree includes: Obtain a recommended quantity threshold, based on the predicted association degree, perform association ranking on the at least two candidate objects, and obtain N candidate objects from the at least two candidate objects after association ranking as the target recommendation objects of the target user; N is a positive integer, and N is greater than or equal to the recommended quantity threshold.

8. The method according to claim 1, wherein The target user and the target recommendation objects are stored in the offline layer; the method further includes: Based on the offline layer, send the target user and the target recommendation objects of the target user to the online layer, and perform associated storage of the target user and the target recommendation objects in the user recommendation library of the online layer; In the online layer, in response to a login operation for the target user, obtain the target recommendation objects associated with the target user from the user recommendation library based on the associated storage relationship, and output the target recommendation objects.

9. The method according to claim 8, characterized in that, The method further includes: In the online layer, in response to a login operation for a to-be-processed user, search for to-be-output recommended objects associated with the to-be-processed user from the user recommendation library based on the associated storage relationship; If there are no to-be-output recommended objects associated with the to-be-processed user in the user recommendation library, obtain default recommended objects, and output the default recommended objects for the to-be-processed user.

10. The method according to claim 5, characterized in that, The method further includes: Obtain at least two training user samples, and obtain sample user features respectively associated with the at least two training user samples from a user feature library; Obtain sample object features associated with each training user sample, and perform model training based on the sample user features and sample object features corresponding to the at least two training user samples respectively to generate a data recommendation model.

11. The method according to claim 10, characterized in that The obtaining at least two training user samples includes: During a training period, obtain at least two candidate users associated with an application; the application is used to output a target recommendation object for the target user; Obtain the corresponding candidate login statuses of the at least two candidate users during the training period, and determine the candidate users whose candidate login statuses include the online status as training user samples; The obtaining of the sample object features associated with each training user sample includes: Count the historical associated objects generated by the training user sample in the online state, and determine the historical object features of the historical associated objects as the sample object features associated with the training user sample.

12. The method according to claim 10, characterized in that, The user feature library includes at least two statistical users and the user features of each statistical user; the at least two statistical users include statistical user j; j is a positive integer, and j is less than or equal to the number of users in the at least two statistical users; The obtaining of the sample user features respectively associated with the at least two training user samples from the user feature library includes: Obtain the training sample hash values respectively corresponding to the at least two training user samples, and generate a sample matching bit array according to the training sample hash values respectively corresponding to the at least two training user samples; Generate the user hash value of the statistical user j, and obtain the user status value of the statistical user j from the sample matching bit array according to the user hash value of the statistical user j; If there is at least one missing status value in the user status value of the statistical user j, it is determined that the statistical user j is not associated with the at least two training user samples; if all the user status values of the statistical user j are valid status values, it is determined that the statistical user j is associated with the at least two training user samples; Determine the user features corresponding to the statistical users associated with the at least two training user samples in the user feature library as the sample user features associated with the at least two training user samples.

13. The method according to claim 10, characterized in that, The at least two training user samples include training user sample k; k is a positive integer, and k is less than or equal to the number of samples of the at least two training user samples; The generating of a data recommendation model based on the sample user features and sample object features respectively corresponding to the at least two training user samples includes: Based on an initial data recommendation model, predict the sample user features and sample object features of the training user sample k, and obtain the sample prediction correlation degree between the training user sample k and the sample object features of the training user sample k; Based on the sample prediction correlation degree and a model loss function, train the initial data recommendation model to generate a data recommendation model.

14. A data recommendation device, characterized in that, The device includes: A state prediction module, configured to obtain the historical login status information of a target user, and predict the predicted login status of the target user within a target time period according to the historical login status information; A feature acquisition module, configured to, if the predicted login status of the target user within the target time period belongs to the online state, search in the user feature library for the target user features associated with the target user, and obtain the object features respectively corresponding to at least two candidate objects; A recommendation prediction module, configured to perform correlation prediction on the object features respectively corresponding to the at least two candidate objects and the target user feature, so as to obtain the prediction correlation degrees of the at least two candidate objects with the target user respectively, and determine a target recommendation object of the target user from the at least two candidate objects based on the prediction correlation degrees; the target recommendation object is used as the recommended content of the target user when it is detected that the actual login status of the target user in the target time period is an online status; The recommendation prediction module is further configured to, if the predicted login status of the target user in the target time period belongs to an offline status, not process the target user, or determine a default recommendation object as the target recommendation object of the target user.

15. A computer device, characterized in that, It includes a processor, a memory, and an input / output interface; The processor is respectively connected to the memory and the input / output interface, wherein the input / output interface is used for receiving and outputting data, the memory is used for storing a computer program, and the processor is used for calling the computer program so that the computer device executes the method according to any one of claims 1-13.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the method according to any one of claims 1-13.

17. A computer program product comprising computer instructions, characterized in that, The computer instructions, when executed by the processor, implement the method according to any one of claims 1-13.

Images