Data processing method, device, storage medium and equipment

Abstract

The embodiment of the present application discloses a data processing method, device, storage medium and equipment. The method includes: obtaining user feature information corresponding to the target user, obtaining the business object to be recommended; the user feature information includes user attribute information corresponding to the target user, and application feature information corresponding to N installed applications associated with the target user; feature encoding the application feature information corresponding to the N installed applications, and obtaining the implicit features corresponding to each installed application; extracting the associated features of the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtaining the attention implicit features between the business object to be recommended and the N installed applications; predicting the user interest of the target user for the business object to be recommended based on the user attribute information and the attention implicit features. The prediction accuracy of the user interest of the target user for the business object to be recommended can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data processing method, device, storage medium and equipment. Background Art

[0002] With the development of Internet technology, network-based content interaction behavior has gradually become a part of people's daily life. In business recommendation services, business objects of interest to users (such as applications, consultations, event information, advertisements, etc.) can be pushed to users based on network interaction to achieve the promotion of business objects.

[0003] Currently, when judging whether a user is interested in a business object, the number of applications of the same type as the business object to be recommended is obtained from the applications that the user has installed in the past, and the user is predicted to be interested in the recommended business object based on the number. The factors used to predict user interest are too single, resulting in the recommendation of business objects that the user is not interested in, and the prediction accuracy is low. Summary of the invention

[0004] The technical problem to be solved by the embodiments of the present application is to provide a data processing method, apparatus, storage medium and device, which can improve the prediction accuracy of the user interest of the target user in the business object to be recommended.

[0005] An embodiment of the present application provides a data processing method, including:

[0006] Obtain user characteristic information corresponding to the target user and obtain the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, wherein the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0007] Feature encoding is performed on the application feature information corresponding to each of the N installed applications to obtain an implicit feature corresponding to each of the installed applications;

[0008] Extract associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0009] According to user attribute information and implicit attention features, the user interest of the target user in the business object to be recommended is predicted.

[0010] The feature encoding is performed on the application feature information corresponding to each of the N installed applications to obtain the implicit features corresponding to each of the installed applications, including:

[0011] Performing vector conversion on application feature information corresponding to each of the N installed applications to obtain an application feature vector corresponding to each of the installed applications;

[0012] The application feature vector corresponding to each installed application is input into the target prediction model. Based on the feature encoding network in the target prediction model, the application feature vector corresponding to each installed application is feature encoded to obtain the implicit features corresponding to each installed application. The target prediction model is used to predict the user's interest in the business object.

[0013] Among them, the feature encoding network includes a self-attention layer and a forward feedback layer;

[0014] Based on the feature encoding network in the target prediction model, feature encoding is performed on the application feature vector corresponding to each installed application to obtain the implicit features corresponding to each installed application, including:

[0015] Based on the attention matrix in the self-attention layer, the application feature vector corresponding to each installed application is extracted with associated features to obtain the target self-attention vector corresponding to each installed application.

[0016] Based on the forward feedback layer, the target self-attention vectors corresponding to each installed application are nonlinearly combined to obtain the implicit features corresponding to each installed application.

[0017] Among them, the self-attention layer includes a self-attention sublayer, a first residual connection sublayer and a first normalization sublayer;

[0018] Based on the attention matrix in the self-attention layer, the application feature vector corresponding to each installed application is extracted with associated features to obtain the target self-attention vector corresponding to each installed application, including:

[0019] Based on the attention matrix in the self-attention sublayer, the application feature vectors corresponding to each installed application are weighted to obtain the initial self-attention vector corresponding to each application feature vector;

[0020] Based on the first residual connection sublayer, residual connection is performed on the initial self-attention vectors corresponding to each application feature vector, so as to obtain the first residual feature vectors corresponding to each application feature vector;

[0021] Based on the first normalization sublayer, the first residual feature vector corresponding to each application feature vector is normalized to obtain the target self-attention vector corresponding to each installed application.

[0022] Among them, based on the attention matrix in the self-attention sublayer, the application feature vectors corresponding to each installed application are weighted to obtain the initial self-attention vectors corresponding to each application feature vector, including:

[0023] Obtain the first matrix, the second matrix, and the third matrix in the self-attention sublayer, and perform weighted processing on the application feature vectors corresponding to each installed application according to the first matrix, the second matrix, and the third matrix, respectively, to obtain the query vector, the key vector, and the value vector corresponding to each installed application;

[0024] For installed applications i The corresponding key vector is transposed to obtain the installed application T i Corresponding transposed key vector, get installed application T i The corresponding query vector and installed applications T i Apply vector dot product between corresponding transposed key vectors; installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0025] According to the dot product of the application vector, the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

[0026] The forward feedback layer includes a fully connected sublayer, a second residual connection sublayer, and a second normalization sublayer;

[0027] Based on the forward feedback layer, the target self-attention vectors corresponding to each installed application are nonlinearly combined to obtain the implicit features corresponding to each installed application, including:

[0028] Based on the fully connected sublayer, convolution is performed on the target self-attention vector corresponding to each installed application to obtain the convolution feature vector corresponding to each installed application.

[0029] Based on the second residual connection sublayer, performing residual connection on the convolution feature vectors corresponding to each installed application, to obtain a second residual feature vector corresponding to each convolution feature vector;

[0030] Based on the second normalization sublayer, the second residual feature vector corresponding to each convolution feature vector is normalized to obtain an implicit feature corresponding to each installed application.

[0031] The implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended are respectively extracted to obtain the attention implicit features between the business object to be recommended and the N installed applications, including:

[0032] Performing vector conversion on the business feature information corresponding to the business object to be recommended, to obtain the business feature vector corresponding to the business object to be recommended;

[0033] Inputting the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, transposing the implicit features corresponding to each installed application to obtain the transposed implicit feature vector corresponding to each installed application; the target prediction model is used to predict the user interest in the business object;

[0034] Based on the first self-attention network, associated features are extracted from the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0035] Among them, based on the first self-attention network, the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended are respectively extracted to obtain the attention implicit features between the business object to be recommended and the N installed applications, including:

[0036] Based on the first self-attention network, get the installed applications T i The first vector dot product between the corresponding transposed implicit feature vector and the business feature vector; Installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0037] Normalizing the dot product of the first vector to obtain a normalized dot product of the first vector;

[0038] According to the normalized first vector dot product, apply T to the installed i The corresponding implicit features are weighted to obtain the installed applications T i Attention application features between the business object to be recommended;

[0039] Based on installed apps T i Attention application features between the business object to be recommended and the N installed applications are determined, and attention implicit features between the business object to be recommended and the N installed applications are determined.

[0040] Among them, predicting the user interest of the target user for the business object to be recommended based on the user attribute information and the implicit characteristics of attention includes:

[0041] Perform vector conversion on the user attribute information to obtain a user attribute vector corresponding to the user attribute information, input the user attribute vector into the perception layer in the target prediction model, and output a global feature vector corresponding to the user attribute vector through the perception layer; the target prediction model is used to predict the user's interest in the business object;

[0042] Inputting the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature into the second self-attention network in the target prediction model, transposing the global feature vector to obtain the transposed global feature vector corresponding to the global feature vector, and obtaining the second vector dot product between the transposed global feature vector and the attention implicit feature vector;

[0043] Normalizing the dot product of the second vector to obtain a normalized dot product of the second vector;

[0044] According to the dot product of the normalized second vector, the global feature vector is weighted to obtain the associated implicit features between the target user and the business object to be recommended;

[0045] The user interest of the target user in the business object to be recommended is predicted based on the associated implicit features between the target user and the business object to be recommended.

[0046] The user attribute vector is input into the perception layer in the target prediction model, and the global feature vector corresponding to the user attribute vector is output through the perception layer, including:

[0047] The user attribute vector is input into the perception layer in the target prediction model, and the user attribute vector is nonlinearly combined based on the bias coefficient and neural matrix in the perception layer to obtain the combined feature vector corresponding to the user attribute vector;

[0048] The combined feature vector is activated through the activation parameters in the perception layer to obtain the global feature vector corresponding to the user attribute vector.

[0049] The data processing method further includes:

[0050] If the user's interest level is greater than or equal to the interest level threshold, the object to be recommended is determined as the target recommended object;

[0051] The target recommended object is sent to the target user terminal corresponding to the target user, so that the target user terminal displays the target recommended object.

[0052] An embodiment of the present application provides a data processing method, including:

[0053] Obtain an initial prediction model, sample user feature information corresponding to the sample user, a sample business object, and a sample user's interest label for the sample business object; the sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, respectively; the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer;

[0054] Using the initial prediction model, feature encoding is performed on the sample application feature information corresponding to each sample installed application, so as to obtain the sample implicit features corresponding to each sample installed application;

[0055] Extracting associated features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object to obtain the sample attention implicit features between the sample business object and the sample installed application;

[0056] Predicting the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit features;

[0057] Determine the loss function corresponding to the initial prediction model based on the predicted user interest and interest label;

[0058] According to the loss function, the parameters of the initial prediction model are adjusted. When the initial prediction model after the parameter adjustment meets the convergence condition, the initial prediction model that meets the convergence condition is used to determine the target prediction model; the target prediction model is used to predict the user interest in the business object.

[0059] An embodiment of the present application provides a data processing device, including:

[0060] The first acquisition module is used to acquire user characteristic information corresponding to the target user and acquire the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0061] A first feature encoding module, used to perform feature encoding on application feature information corresponding to N installed applications respectively, to obtain implicit features corresponding to each installed application respectively;

[0062] A first correlation feature extraction module is used to extract correlation features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0063] The first prediction module is used to predict the user interest of the target user for the business object to be recommended based on the user attribute information and the implicit feature of attention.

[0064] Wherein, the first feature encoding module includes:

[0065] A first vector conversion unit, configured to perform vector conversion on application feature information corresponding to each of the N installed applications to obtain an application feature vector corresponding to each of the installed applications;

[0066] The feature encoding unit is used to input the application feature vector corresponding to each installed application into the target prediction model, and based on the feature encoding network in the target prediction model, the application feature vector corresponding to each installed application is feature encoded to obtain the implicit features corresponding to each installed application; the target prediction model is used to predict the user's interest in the business object.

[0067] Among them, the feature encoding network includes a self-attention layer and a forward feedback layer;

[0068] The feature encoding unit includes:

[0069] A correlation feature extraction subunit is used to extract correlation features from the application feature vectors corresponding to each installed application based on the attention matrix in the self-attention layer, so as to obtain a target self-attention vector corresponding to each installed application;

[0070] The first nonlinear combination subunit is used to perform nonlinear combination on the target self-attention vectors corresponding to each installed application based on the forward feedback layer to obtain the implicit features corresponding to each installed application.

[0071] Among them, the self-attention layer includes a self-attention sublayer, a first residual connection sublayer and a first normalization sublayer;

[0072] The associated feature extraction subunit is specifically used for:

[0073] Based on the attention matrix in the self-attention sublayer, the application feature vectors corresponding to each installed application are weighted to obtain the initial self-attention vector corresponding to each application feature vector;

[0074] Based on the first residual connection sublayer, residual connection is performed on the initial self-attention vectors corresponding to each application feature vector, so as to obtain the first residual feature vectors corresponding to each application feature vector;

[0075] Based on the first normalization sublayer, the first residual feature vector corresponding to each application feature vector is normalized to obtain the target self-attention vector corresponding to each installed application.

[0076] The associated feature extraction subunit is also specifically used for:

[0077] Obtain the first matrix, the second matrix, and the third matrix in the self-attention sublayer, and perform weighted processing on the application feature vectors corresponding to each installed application according to the first matrix, the second matrix, and the third matrix, respectively, to obtain the query vector, the key vector, and the value vector corresponding to each installed application;

[0078] For installed applications i The corresponding key vector is transposed to obtain the installed application T i Corresponding transposed key vector, get installed application T i The corresponding query vector and installed applications T i Apply vector dot product between corresponding transposed key vectors; installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0079] According to the dot product of the application vector, the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

[0080] The forward feedback layer includes a fully connected sublayer, a second residual connection sublayer, and a second normalization sublayer;

[0081] The first nonlinear combination subunit is specifically used for:

[0082] Based on the fully connected sublayer, convolution is performed on the target self-attention vector corresponding to each installed application to obtain the convolution feature vector corresponding to each installed application.

[0083] Based on the second residual connection sublayer, performing residual connection on the convolution feature vectors corresponding to each installed application, to obtain a second residual feature vector corresponding to each convolution feature vector;

[0084] Based on the second normalization sublayer, the second residual feature vector corresponding to each convolution feature vector is normalized to obtain an implicit feature corresponding to each installed application.

[0085] Wherein, the first correlation feature extraction module includes:

[0086] A second vector conversion unit is used to perform vector conversion on the service feature information corresponding to the service object to be recommended, so as to obtain a service feature vector corresponding to the service object to be recommended;

[0087] A transposition processing unit, used to input the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, and perform transposition processing on the implicit features corresponding to each installed application to obtain a transposed implicit feature vector corresponding to each installed application; the target prediction model is used to predict the user interest in the business object;

[0088] The associated feature extraction unit is used to extract associated features from the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended based on the first self-attention network, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0089] Wherein, the associated feature extraction unit includes:

[0090] Get subunit, used to get installed applications T based on the first self-attention network i The first vector dot product between the corresponding transposed implicit feature vector and the business feature vector; Installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0091] A normalization subunit, used for normalizing the first vector dot product to obtain a normalized first vector dot product;

[0092] The weighted processing subunit is used to apply T to the installed i The corresponding implicit features are weighted to obtain the installed applications T i Attention application features between the business object to be recommended;

[0093] Determine the subunit for installing the application T i Attention application features between the business object to be recommended and the N installed applications are determined, and attention implicit features between the business object to be recommended and the N installed applications are determined.

[0094] Wherein, the first prediction module includes:

[0095] The output unit is used to perform vector conversion on the user attribute information to obtain a user attribute vector corresponding to the user attribute information, input the user attribute vector into the perception layer in the target prediction model, and output a global feature vector corresponding to the user attribute vector through the perception layer; the target prediction model is used to predict the user interest in the business object;

[0096] An acquisition unit, used for inputting the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature into the second self-attention network in the target prediction model, transposing the global feature vector to obtain the transposed global feature vector corresponding to the global feature vector, and obtaining a second vector dot product between the transposed global feature vector and the attention implicit feature vector;

[0097] A normalization processing unit, used for performing normalization processing on the second vector dot product to obtain a normalized second vector dot product;

[0098] A weighted processing unit, used to perform weighted processing on the global feature vector according to the dot product of the normalized second vector, so as to obtain an associated implicit feature between the target user and the business object to be recommended;

[0099] The prediction unit is used to predict the user interest of the target user in the business object to be recommended based on the associated implicit features between the target user and the business object to be recommended.

[0100] The output unit includes:

[0101] The second nonlinear combination subunit is used to input the user attribute vector into the perception layer in the target prediction model, and perform nonlinear combination on the user attribute vector based on the bias coefficient and the neural matrix in the perception layer to obtain a combined feature vector corresponding to the user attribute vector;

[0102] The activation processing unit is used to activate the combined feature vector through the activation parameters in the perception layer to obtain the global feature vector corresponding to the user attribute vector.

[0103] The data processing device further comprises:

[0104] A first determination module, configured to determine the object to be recommended as a target recommendation object if the user's interest level is greater than or equal to an interest level threshold;

[0105] The sending module is used to send the target recommended object to the target user terminal corresponding to the target user, so that the target user terminal displays the target recommended object.

[0106] An embodiment of the present application provides a data processing device, including:

[0107] The second acquisition module is used to acquire the initial prediction model, sample user feature information corresponding to the sample user, the sample business object, and the sample user's interest label for the sample business object; the sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, respectively; the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer;

[0108] The second feature encoding module is used to use the initial prediction model to perform feature encoding on the sample application feature information corresponding to each sample installed application, so as to obtain the sample implicit feature corresponding to each sample installed application;

[0109] A second association extraction module is used to extract association features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object, so as to obtain the sample attention implicit features between the sample business object and the sample installed application;

[0110] The second prediction module is used to predict the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit feature;

[0111] A second determination module is used to determine the loss function corresponding to the initial prediction model according to the predicted user interest and interest label;

[0112] The third determination module is used to adjust the parameters of the initial prediction model according to the loss function. When the initial prediction model after parameter adjustment meets the convergence condition, the initial prediction model that meets the convergence condition is used to determine the target prediction model; the target prediction model is used to predict the user's interest in the business object.

[0113] On one hand, the present application provides a computer device, including: a processor and a memory;

[0114] The memory is used to store the computer program, and the processor is used to call the computer program to perform the following steps:

[0115] Obtain user characteristic information corresponding to the target user and obtain the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, wherein the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0116] Feature encoding is performed on the application feature information corresponding to each of the N installed applications to obtain an implicit feature corresponding to each of the installed applications;

[0117] Extract associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0118] According to user attribute information and implicit attention features, the user interest of the target user in the business object to be recommended is predicted.

[0119] In one aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the computer program performs the following steps:

[0120] Obtain user characteristic information corresponding to the target user and obtain the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, wherein the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0121] Feature encoding is performed on the application feature information corresponding to each of the N installed applications to obtain an implicit feature corresponding to each of the installed applications;

[0122] Extract associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0123] According to user attribute information and implicit attention features, the user interest of the target user in the business object to be recommended is predicted.

[0124] In one aspect, the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a 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 method in the above aspect.

[0125] In an embodiment of the present application, the business object to be recommended is obtained by obtaining user feature information corresponding to the target user; the user feature information includes user attribute information corresponding to the target user, and application feature information corresponding to N installed applications associated with the target user, and feature encoding is performed on the application feature information corresponding to the N installed applications to obtain implicit features corresponding to each installed application. Through the implicit features corresponding to each installed application, the associated feature information between each installed application can be better extracted; the associated features are extracted from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and the attention implicit features between the business object to be recommended and the N installed applications are obtained. The attention implicit features are obtained by self-attention mechanism. The importance of different features can be better extracted through the self-attention mechanism, so as to focus on extracting features with greater importance, which can increase the accuracy of feature extraction for the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the attention implicit features can better characterize the implicit relationship (such as association relationship) between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended; according to the user attribute information and the attention implicit features, the user interest of the target user for the business object to be recommended is predicted, and the user interest of the target user for the business object to be recommended is predicted by using the attention implicit features and user attribute information that contain the implicit relationship between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the user interest can be predicted more accurately. In the present application, the prediction accuracy of the user interest of the target user for the business object to be recommended can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0127] Figure 1 It is a schematic diagram of the architecture of a data processing system provided in an embodiment of the present application;

[0128] Figure 2 This is a schematic diagram of an application scenario of data processing provided by an embodiment of the present application;

[0129] Figure 3 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0130] Figure 4 It is a structural diagram of a feature encoding network provided in an embodiment of the present application;

[0131] Figure 5 is a schematic diagram of a method for obtaining implicit features of attention provided in an embodiment of the present application;

[0132] Figure 6 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0133] Figure 7 is a schematic diagram of a method for obtaining user interest provided in an embodiment of the present application;

[0134] Figure 8 This is an application scenario diagram for recommending business objects based on user interest provided by an embodiment of the present application;

[0135] Fig. 9 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0136] Fig.10 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0137] Fig.11 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0138] Fig.12 is a structural schematic diagram of a computer device provided in an embodiment of the present application;

[0139] Fig.13 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0140] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0141] This application relates to artificial intelligence (AI) technology, blockchain (BlockChain) technology and cloud technology. Artificial intelligence is the theory, method, technology and application system of using digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is 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. Artificial intelligence technology is a comprehensive discipline that involves a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning. This application specifically relates to machine learning (ML) under artificial intelligence technology.

[0142] Among them, machine learning (ML) is a multi-field interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning. In this application, a target prediction model with an attention network structure or a self-attention network structure can be trained and generated through machine learning. The user's interest in a business object is predicted by the target prediction model, and the prediction accuracy of the user's interest can be improved.

[0143] The present application relates to cloud storage under cloud technology. Cloud storage is a new concept extended and developed from the concept of cloud computing. A distributed cloud storage system (hereinafter referred to as storage system) refers to a storage system that uses cluster applications, grid technology, and distributed storage file systems to bring together a large number of storage devices of various types (storage devices are also called storage nodes) in the network to work together through application software or application interfaces to provide data storage and business access functions to the outside world. At present, the storage method of the storage system is: create a logical volume. When creating a logical volume, a physical storage space is allocated to each logical volume. The physical storage space may be composed of disks of a storage device or several storage devices. The client stores data on a logical volume, that is, the data is stored on the file system. The file system divides the data into many parts, each of which is an object. The object not only contains data but also contains additional information such as data identification (ID entity, ID). The file system writes each object into the physical storage space of the logical volume, and the file system records the storage location information of each object, so that when the client requests to access the data, the file system can allow the client to access the data according to the storage location information of each object.

[0144] The process of the storage system allocating physical storage space to a logical volume is as follows: based on the estimated capacity of the objects stored in the logical volume (this estimate often has a large margin relative to the actual capacity of the objects to be stored) and the grouping of the Redundant Array of Independent Disks (RAID), the physical storage space is divided into stripes in advance. A logical volume can be understood as a stripe, thereby allocating physical storage space to the logical volume. In the present application, user data corresponding to different user terminals (i.e., registered users owned by each user terminal) can be stored on different logical volumes respectively, that is, user data held by each user terminal can be stored on the file system; for the user data held by any user terminal, the file system can divide it into many parts, each part is an object (i.e., user), and the object can not only contain user feature information corresponding to the user (i.e., user attribute information and application feature information corresponding to N installed applications), but also contain user identification information; the file system writes each object into the physical storage space of the logical volume respectively, and the file system records the storage location information of each object; in this way, when determining the user's interest in a business object, the server or user terminal can request access to the user data in the file system, and the file system can allow the server or user terminal to access the user data based on the storage location information of the object.

[0145] Blockchain is a new application model of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks (also called blocks) generated by cryptography. Each data block is linked by random hashing (also called hash algorithm). The latter block contains the encrypted hash of the previous block, the corresponding timestamp and transaction data (usually represented by the hash value calculated by the Merkle tree algorithm). This design makes the block content difficult to tamper with. The distributed ledger connected by blockchain technology allows both parties to effectively record transactions and can permanently check the transaction. All service devices involved in this application can be blockchain nodes belonging to the same blockchain system. In the process of recommending business objects to users, user recommendation records can be stored on the blockchain to ensure the traceability of user recommendation records.

[0146] See Figure 1 , Figure 1 Schematic diagram of a data processing system provided in an embodiment of the present application. Figure 1 As shown, the data processing system may include a server 10 and a user terminal cluster. The user terminal cluster may include one or more user terminals, and the number of user terminals is not limited here. Figure 1 As shown, it may specifically include user terminal 100a, user terminal 100b, user terminal 100c, ..., user terminal 100n. Figure 1 As shown, the user terminal 100a, the user terminal 100b, the user terminal 100c, ..., the user terminal 100n can respectively establish a network connection with the server 10, so that each user terminal can exchange data with the server 10 through the network connection.

[0147] Each user terminal in the user terminal cluster may include: a smart phone, a tablet computer, a laptop computer, a desktop computer, a wearable device, a smart home, a head-mounted device, and other smart terminals with trajectory recognition. Figure 1 Each user terminal in the user terminal cluster shown in FIG. 1 may be installed with a target application (i.e., an application client). When the application client runs in each user terminal, it may be respectively connected to the above-mentioned Figure 1 The servers 10 shown here interact with each other.

[0148] Among them, Figure 1As shown, the server 10 can be an independent physical server, a server cluster or a 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, CDN, as well as big data and artificial intelligence platforms.

[0149] For ease of understanding, the embodiments of the present application can be Figure 1 A user terminal is selected from the multiple user terminals shown as the target user terminal, and the target user terminal may include: a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, and other smart terminals with data processing functions. For example, for ease of understanding, the embodiment of the present application may Figure 1 The user terminal 100a shown is used as a target user terminal. The user terminal 100a may refer to a target user terminal corresponding to a target user. If it is currently necessary to predict the user interest of user A for a business object Q, user A may be determined as a target user, and business object Q may be determined as a to-be-recommended object. The user terminal 100a may send user feature information corresponding to the target user (i.e., the user for whom the user terminal 100a currently needs to predict the user interest for the to-be-recommended object, such as user A) and the to-be-recommended business object (such as business object Q) to the server 10. The user feature information includes user attribute information corresponding to the target user, and application feature information corresponding to N installed applications associated with the target user. The N installed applications have the same attribute type as the to-be-recommended business object, and N is a positive integer, such as N may be 1, 2, 3, etc. After receiving the user feature information and the to-be-recommended business object sent by the user terminal 100a, the server 10 may perform feature encoding on the application feature information corresponding to the N installed applications, and obtain implicit features corresponding to each installed application. After the server 10 obtains the implicit features corresponding to each installed application, it can extract the associated features of the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications. The server 10 can predict the user interest of the target user for the business object to be recommended based on the user attribute information and the attention implicit features. After the server obtains the user interest of the target user for the business object to be recommended, it can send the user interest to the user terminal 100a. After the user terminal 100a receives the user interest, it can provide business services for the target user based on the user interest, such as pushing the business object to be recommended to the target user, or pushing the relevant information (consulting information or activity information, etc.) corresponding to the business object to be recommended to the target user.

[0150] like Figure 2 As shown, Figure 2 is a schematic diagram of an application scenario of data processing provided by an embodiment of the present application, such as Figure 2 As shown, the business object to be recommended may refer to a game application, and the N installed applications having the same attributes as the business object to be recommended may refer to installed game applications that the target user has installed, where N is a positive integer, such as N may be 1, 2, 3, etc. Figure 2 As shown, after user 20a clicks on the application manager in the desktop interface 20c of user terminal 20b, user terminal 20b jumps to the application download interface, and user 20a can select the category recommendation option. When user 20a clicks on the category recommendation touch key, user terminal 20b can respond to the trigger operation of user 20a on the category recommendation touch key and display the display interface 20d. User 20a can select the type of application to be downloaded in the display interface 20d, such as user 20a can select the game type. When user 20a clicks on the game type in the display interface 20d, user terminal 20b can respond to the trigger operation of user 20a on the game application, jump to the interface, and display the display interface 20e, prompting user 20a that the recommended game application is currently being generated. In the process of obtaining the recommended game applications, the user terminal 20b can determine the target user as the user 20a, obtain the user characteristic information corresponding to the user 20a, and obtain the N installed game applications that the user 20a has installed, and generate an acquisition request for the user interest of the target user for the game applications to be recommended according to the user characteristic information corresponding to the user 20a, the N installed game applications that the user 20a has installed, and the set of game applications to be recommended, and send the acquisition request to the server 20f. After receiving the acquisition request sent by the user terminal 20b, the server 20f can predict the user interest of the user 20a for each game application to be recommended in the set of game applications to be recommended by using the target prediction model for predicting the user interest of the user for the business object. After obtaining the user interest of the user 20a for each game application to be recommended in the set of game applications to be recommended, the server 20f can send the user interest corresponding to each game application to be recommended to the user terminal 20b. After the user terminal 20b receives the user interest level of each game application to be recommended in the set of game applications to be recommended from the user 20a, the user interest level corresponding to each game application to be recommended can be sorted to obtain the sorted user interest level. The user terminal 20b can determine the top X game applications in the sorting as target recommended game applications, where X is a positive integer, such as X can be 1, 2, 3..., and display the target recommended game applications in the display page 20g, such as game application 1, game application 2, game application 3, etc. The user 20a can click to install each target recommended game application, thereby accurately recommending business objects to users and avoiding unnecessary troubles to users.

[0151] See also Figure 3 , Figure 3 is a flow chart of a data processing method provided in an embodiment of the present application. The data processing method can be executed by a computer device, which can be a server (such as the above-mentioned Figure 1 Server 10 in the example above) or a user terminal (such as the one described above) Figure 1 Any user terminal in a user terminal cluster), or a system consisting of a server and a user terminal, which is not limited in this application. Figure 3 As shown, the data processing method may include steps S101-S104.

[0152] S101, obtaining user feature information corresponding to a target user, and obtaining a business object to be recommended.

[0153] Specifically, in a business promotion scenario, a computer device can obtain the user interest of a target user (i.e., a user who currently needs to recommend a business object) for the business object to be recommended, and determine whether it is necessary to place the business object to be recommended to the target user based on the user interest, so as to achieve accurate push of the business object. At the same time, when the target user is not interested in the recommended business object, no push is performed to avoid interference with the target user. The computer device can obtain user feature information corresponding to the target user, and obtain the business object to be recommended. The user feature information corresponding to the target user includes user attribute information corresponding to the target user, and application feature information corresponding to N installed applications associated with the target user, respectively. The N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer. Among them, the user attribute information corresponding to the target user may refer to the user's age, user gender, user occupation, user education, user consumption ability, user payment habits, etc., and the application feature information corresponding to the installed application may refer to the application type, the first installation time of the application, the application price, the application update time, etc. The business object to be recommended may refer to applications (such as game applications, video applications, shopping applications, etc.), information (such as game information, video information, etc.), gift packages (such as game gift packages, video gift packages (such as video membership gift packages, game skin gift packages, etc.)), promotional activities (such as game activities, music activities, etc.), etc. The N installed applications have the same attribute type as the business object to be recommended, and the attribute type may be a business type, an application type, a usage type, a business function type, etc. For example, when the business object to be recommended is a game application, the N installed applications may refer to installed game applications that have been installed by the target user. For example, when the business object to be recommended is a game gift package, the N installed applications may also refer to installed game applications that have been installed by the target user, that is, both the game gift package and the game application are of the game type. For example, when the business object to be recommended is a video application, the N installed applications may refer to installed video applications that have been installed by the target user. The N installed applications are applications that the user has installed, and contain interest factors that the user is interested in. The user interest of the target user in the business object to be recommended can be predicted based on the user attribute information corresponding to the target user, the application feature information corresponding to each installed application, and the business feature information corresponding to the business object to be recommended.

[0154] S102, feature encoding is performed on application feature information corresponding to N installed applications to obtain implicit features corresponding to each installed application.

[0155] Specifically, the computer device can perform feature encoding on the application feature information corresponding to N installed applications respectively to obtain the implicit features corresponding to each installed application respectively, that is, perform global feature extraction on the application feature information corresponding to each installed application to obtain the implicit features corresponding to each installed application, that is, the implicit relationship between the application feature information of each installed application can be obtained, such as the association relationship between the installed applications.

[0156] Optionally, the specific manner in which the computer device obtains the implicit features corresponding to each installed application may include: performing vector conversion on the application feature information corresponding to each installed application in the N installed applications to obtain the application feature vector corresponding to each installed application. The application feature vector corresponding to each installed application is input into the target prediction model, and based on the feature encoding network in the target prediction model, the application feature vector corresponding to each installed application is feature encoded to obtain the implicit features corresponding to each installed application; the target prediction model is used to predict the user's interest in the business object.

[0157] Specifically, the computer device can perform vector conversion on the application feature information corresponding to each installed application in the N installed applications to obtain the application feature vector corresponding to each installed application. The application feature vector refers to an object that has both size and direction and satisfies the parallelogram law, and can be used to characterize the application feature information corresponding to the installed application. Among them, the computer device can use vector conversion methods such as Embedding (embedding network, which can convert discrete variables into continuous vectors. Converting into continuous vectors can not only reduce the spatial dimension of discrete variables, but also meaningfully represent discrete variables), One-hot encoding (One-Hot Encoding, also known as unique hot code, used to convert discrete variables into binary vectors), Word2vec encoding (a vector conversion method used to map each feature to a vector, which can be used to represent the relationship between features) to perform vector conversion on the application feature information corresponding to each installed application. For example, when a computer device uses Embedding to perform vector conversion on the application feature information corresponding to each installed application, the application feature information corresponding to each installed application can be mapped to a vector on the real number field in a low-dimensional dense continuous vector space, similar application feature information in the application feature information can be found, and similar feature information can be mapped using vectors with similar distances, that is, application feature information with similar vector distances have similar meanings. After the computer device obtains the application feature vector corresponding to each installed application, the application feature vector corresponding to each installed application can be input into the target prediction model, and the application feature vector corresponding to each installed application can be feature encoded through the feature encoding network in the target prediction model to obtain the implicit features corresponding to each installed application. The target prediction model is used to predict the user's interest in the business object. The feature encoding network in the target prediction model converts the vector to be converted into a target vector in the target form, that is, the application feature vector corresponding to each installed application is converted into a corresponding implicit vector.

[0158] Optionally, the feature encoding network includes a self-attention layer and a forward feedback layer, and the specific manner in which the computer device obtains the implicit features corresponding to each installed application may include: based on the attention matrix in the self-attention layer, performing associated feature extraction on the application feature vector corresponding to each installed application, to obtain a target self-attention vector corresponding to each installed application; based on the forward feedback layer, performing nonlinear combination on the target self-attention vector corresponding to each installed application, to obtain the implicit features corresponding to each installed application.

[0159] Specifically, the computer device may input the application feature vector corresponding to each installed application into the self-attention layer in the feature encoding network, and based on the attention matrix in the self-attention layer, perform associated feature extraction on the application feature vector corresponding to each installed application, and obtain the target self-attention vector corresponding to each installed application. Based on the forward feedback layer, perform nonlinear combination on the target self-attention vector corresponding to each installed application, and obtain the implicit features corresponding to each installed application. Among them, the feature encoding network may refer to the Transformer network structure. The Transformer neural network is a neural network based on the encoder (encoder)-decoder (decoder) structure, which is used to convert the vector to be converted into a target vector in the target form, and is composed of a self-attention self-attention network and a feedforward forward network. The self-attention layer may refer to the self-attention in the Transformer neural network. The self-attention refers to the self-attention mechanism, which is used to calculate the matching degree between the current input sequence and the output vector. The input sequence with a high matching degree has a larger corresponding matching degree weight, so as to focus on important feature information (i.e., focus on extracting important feature information) while ignoring other unimportant factors. The forward feedback layer may refer to a feed forward network, which is also called a multi-layer perceptron, and is composed of multiple layers of continuous non-linear functions and multiple layers of perceptrons (discontinuous non-linear functions), and is used to enhance the features of feature vectors.

[0160] Optionally, the self-attention layer includes a self-attention sublayer, a first residual connection sublayer, and a first normalization sublayer, and the specific method for the computer device to obtain the target self-attention vector corresponding to each installed application may include: based on the attention matrix in the self-attention sublayer, weighting the application feature vector corresponding to each installed application to obtain the initial self-attention vector corresponding to each application feature vector. Based on the first residual connection sublayer, residual connection is performed on the initial self-attention vector corresponding to each application feature vector to obtain the first residual feature vector corresponding to each application feature vector, and based on the first normalization sublayer, the first residual feature vector corresponding to each application feature vector is normalized to obtain the target self-attention vector corresponding to each installed application.

[0161] Specifically, the computer device can perform weighted processing on the application feature vectors corresponding to each installed application based on the attention matrix in the self-attention sublayer to obtain the initial self-attention vectors corresponding to each application feature vector, that is, through the self-attention sublayer, the application feature vectors corresponding to each installed application are extracted with associated features, so that the application feature information corresponding to each installed application contains the application feature information corresponding to other installed applications. For example, N installed applications include installed application 1, installed application 2, and installed application 3. Based on the attention matrix in the self-attention sublayer, the application feature vectors corresponding to installed application 1, installed application 2, and installed application 3 are weighted, and the initial self-attention vectors corresponding to the three installed applications are obtained. The initial self-attention vectors corresponding to each installed application contain the application feature information corresponding to other installed applications, such as the initial self-attention vector corresponding to installed application 1 contains the application feature information corresponding to installed application 2 and installed application 3. Based on the first residual connection sublayer in the self-attention layer, the initial self-attention vector corresponding to each installed application is convoluted to obtain the convolution self-attention vector corresponding to each installed application, and the vector sum corresponding to the application feature vector corresponding to each installed application and the convolution self-attention vector corresponding to each installed application is obtained, and the vector sum is determined as the first residual feature vector corresponding to each application feature vector. The role of performing residual connection on the initial self-attention vector corresponding to each installed application is to prevent the gradient of the model from disappearing, which can accelerate the convergence of the model. Specifically, the number of the first residual connection sublayer can be one or more, and the dimension of the input data of each first residual connection sublayer is the same as the dimension of the output data. Therefore, based on the first residual connection sublayer (one or more) in the self-attention layer, after the initial self-attention vector of size C×H×W is residually connected, the size of the residual feature vector obtained is still C×H×W.

[0162] Residual connection can refer to the case where after two consecutive convolution operations, the scope of the activation function is the sum of the output after the convolution operation and the input before the convolution operation. The calculation formula of the residual connection is the following formula (1):

[0163]

[0164] Among them, R out is the output of the first residual connection sublayer (i.e., the first residual feature vector); x is the input of the first residual connection sublayer; W1, W2 represent the weights in the first convolution operation and the second convolution operation, respectively; σ represents the activation function.

[0165] Specifically, after the computer device obtains the first residual feature vector corresponding to each installed application, it can normalize the first residual feature vector corresponding to each installed application based on the first normalization sublayer in the self-attention layer, that is, perform nonlinear combination on the first residual feature vector corresponding to each installed application to obtain the target self-attention vector corresponding to each installed application. The function of the first normalization layer is also to prevent the gradient of the model from disappearing, which can accelerate the convergence of the model.

[0166] Optionally, the specific method for the computer device to obtain the initial self-attention vector corresponding to each application feature vector may include: obtaining the first matrix, the second matrix and the third matrix in the self-attention sublayer, and performing weighted processing on the application feature vector corresponding to each installed application according to the first matrix, the second matrix and the third matrix, respectively, to obtain the query vector, the key vector and the value vector corresponding to each installed application. i The corresponding key vector is transposed to obtain the installed application T i Corresponding transposed key vector, get installed application T i The corresponding query vector and installed applications T i Apply vector dot product between corresponding transposed key vectors; installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N, and according to the dot product of the application vector, the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

[0167] Specifically, the computer device may obtain the first matrix, the second matrix, and the third matrix in the self-attention sublayer, where the first matrix may refer to the query weight matrix W Q , the second matrix can refer to the key weight matrix W K , the third matrix can be the value weight matrix W V . Using this W Q Query weight matrix, key weight matrix W K , value weight matrix W V The application feature vectors corresponding to each installed application are weighted (i.e., linearly mapped) to obtain the query vector, key vector, and value vector corresponding to each installed application, i.e., the query weight matrix W is used. Q The application feature vectors corresponding to each installed application are weighted to obtain the query vectors corresponding to each installed application. The key weight matrix W is used K The application feature vectors corresponding to each installed application are weighted to obtain the key vectors corresponding to each installed application, using the value weight matrix WV Perform weighted processing on the application feature vectors corresponding to each installed application to obtain the value vectors corresponding to each installed application. i , i is a positive integer less than or equal to N, which can be used for installed applications T i The corresponding key vector is transposed to obtain the installed application T i Corresponding transposed key vector, get installed application T i The corresponding query vector corresponds to the installed application T i Apply vector dot products between the transposed key vectors. The installed application T i The corresponding query vector corresponds to the installed application T i The dot product of the application vector between the transposed key vectors of , which can be called the attention matrix in the self-attention sublayer, can be used to represent the installed applications T i With the exception of installed applications i The larger the dot product of the application vector, the greater the similarity between the vectors; the smaller the dot product of the application vector, the smaller the similarity between the vectors. i After the corresponding application vector dot product, the application vector dot product can be used as a weight matrix to calculate the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

[0168] Optionally, the forward feedback layer includes a fully connected sublayer, a second residual connection sublayer, and a second normalization sublayer, and the computer device obtains the implicit features corresponding to each installed application, which may include: based on the fully connected sublayer, convolution processing is performed on the target self-attention vector corresponding to each installed application, to obtain the convolution feature vector corresponding to each installed application. Based on the second residual connection sublayer, residual connection is performed on the convolution feature vector corresponding to each installed application, to obtain the second residual feature vector corresponding to each convolution feature vector, and based on the second normalization sublayer, the second residual feature vector corresponding to each convolution feature vector is normalized to obtain the implicit features corresponding to each installed application.

[0169] Specifically, the computer device can input the target self-attention vector corresponding to each installed application into the forward feedback layer, and perform convolution processing on the target self-attention vector corresponding to each installed application through the fully connected sublayer in the forward feedback layer to obtain the convolution feature vector corresponding to each installed application. The fully connected sublayer may include one or more fully connected processing layers for performing convolution processing on the target self-attention vector corresponding to each installed application, that is, performing feature extraction. An activation function can be set after each fully connected processing layer to activate the output of each fully connected processing layer, thereby increasing the nonlinear expression ability of the forward feedback layer. After the computer device obtains the convolution feature vector corresponding to each installed application, it can perform residual connection on the convolution feature vector corresponding to each installed application based on the second residual connection sublayer to obtain the second residual feature vector corresponding to each convolution feature vector. The specific residual connection content can refer to the above description of formula (1), and the embodiment of the present application will not be repeated here. After the computer device obtains the second residual feature vector corresponding to each convolution feature vector, it can normalize the second residual feature vector corresponding to each convolution feature vector based on the second normalization sublayer to obtain the implicit features corresponding to each installed application.

[0170] like Figure 4 As shown, Figure 4 is a schematic diagram of the structure of a feature encoding network provided in an embodiment of the present application, such as Figure 4As shown, the feature encoding network may include a self-attention layer 40a and a forward feedback layer 40b. Among them, the self-attention layer 40a includes a self-attention sublayer 40a1, a first residual connection sublayer 40a2 and a first normalization sublayer 40a3. The self-attention layer 40a is used to extract associated features from the input feature vector (such as the application feature vector corresponding to each installed application). The global feature vector (such as the application feature vector corresponding to N installed applications) can be fused into each local feature vector (the application features corresponding to each installed application are adjacent to each other) to extract the correlation between each local feature vector. The self-attention sublayer 40a1 can use an attention matrix to perform weighted processing on the application feature vector corresponding to each installed application, and obtain the initial self-attention vector corresponding to each application feature vector. The first residual connection sublayer 40a2 can perform residual connection on the initial self-attention vector corresponding to each application feature vector to obtain the first residual feature vector corresponding to each application feature vector, that is, after convolution processing on the initial self-attention vector, the initial self-attention vector after convolution processing is added to the application feature vector to obtain the first residual feature vector. The first normalization sublayer 40a3 can perform normalization processing on the first residual feature vector corresponding to each application feature vector to obtain the target self-attention vector corresponding to each installed application. The forward feedback layer 40b includes a fully connected sublayer 40b1, a second residual connection sublayer 40b2 and a second normalization sublayer 40b3. The forward feedback layer 40b can be used to perform nonlinear combination on the input feature vector. Enhance the feature information of the input feature vector. The target self-attention vector corresponding to each installed application can be convolved through the fully connected sublayer 40b1 to obtain the convolution feature vector corresponding to each installed application. The convolution feature vectors corresponding to each installed application can be residually connected through the second residual connection sublayer 40b2 to obtain the second residual feature vectors corresponding to each convolution feature vector. The second residual feature vectors corresponding to each convolution feature vector can be normalized through the second normalization sublayer 40b3 to obtain the implicit features corresponding to each installed application. The specific processing process of the feature encoding network can refer to the content described in step S102, and the embodiment of the present application will not be repeated here.

[0171] S103, extracting associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, to obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0172] Specifically, after the computer device obtains the implicit features corresponding to each installed application, it can perform associated feature extraction on the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0173] Optionally, the specific method for the computer device to obtain the implicit feature of attention between the business object to be recommended and the N installed applications may include: performing vector conversion on the business feature information corresponding to the business object to be recommended to obtain the business feature vector corresponding to the business object to be recommended, inputting the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, transposing the implicit features corresponding to each installed application to obtain the transposed implicit feature vector corresponding to each installed application, and the target prediction model is used to predict the user's interest in the business object. Based on the first self-attention network, the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended are subjected to correlation feature extraction to obtain the implicit feature of attention between the business object to be recommended and the N installed applications.

[0174] Specifically, the computer device can perform vector conversion on the business feature information corresponding to the recommended business object to obtain the business feature vector corresponding to the recommended business object, wherein the vector conversion method can be Embedding (embedding network, which can convert discrete variables into continuous vectors. Converting into continuous vectors can not only reduce the spatial dimension of discrete variables, but also meaningfully represent discrete variables), One-hot encoding (One-Hot Encoding, also known as unique hot code, used to convert discrete variables into binary vectors), Word2vec encoding (a vector conversion method used to map each feature to a vector, which can be used to represent the relationship between features) and other vector conversion methods. For specific content, please refer to the above-mentioned content of vector conversion for each installed application, and the embodiments of this application will not be repeated here. The computer device can input the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, and transpose the implicit features corresponding to each installed application to obtain the transposed implicit feature vector corresponding to each installed application. The target prediction model is used to predict the user's interest in the business object. Based on the first self-attention network, the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended are extracted to obtain the attention implicit features between the business object to be recommended and the N installed applications. Through the first self-attention network, the implicit features corresponding to each installed application can be combined with the business object to be recommended to obtain the similarity between each installed application and the business object to be recommended.

[0175] Optionally, the specific method for the computer to obtain the attention implicit feature between the to-be-recommended business object and the N installed applications may include: based on the first self-attention network, obtaining the installed applications T i The first vector dot product between the corresponding transposed implicit feature vector and the business feature vector, the installed application Ti belongs to N installed applications, i is a positive integer less than or equal to N. The first vector dot product is normalized to obtain the normalized first vector dot product, and the installed application T is calculated based on the normalized first vector dot product. i The corresponding implicit features are weighted to obtain the installed applications T i The attention application features between the recommended business object and the installed application T i Attention application features between the business object to be recommended and the N installed applications are determined, and attention implicit features between the business object to be recommended and the N installed applications are determined.

[0176] Specifically, the computer device can obtain the installed application T based on the first self-attention network. i The first vector dot product between the corresponding transposed implicit feature vector and the business feature vector, that is, the transposed implicit feature vector corresponding to an installed application is dot-producted with the business feature vector to obtain a first vector dot product, which can also be called a scalar product or a dot product. The dot product of two unit vectors obtains the cosine value of the angle between the two vectors. Through the vector dot product between the two vectors, the similarity between the two vectors can be known. Therefore, the first vector dot product between each installed application and the business object to be recommended can be used to obtain the familiarity between each installed application and the business object to be recommended. The first vector dot product is normalized to obtain the normalized first vector dot product. According to the normalized first vector dot product, the installed application T i The corresponding implicit features are weighted to obtain the installed applications T i The attention application feature between the recommended business object and the application object, that is, the attention application feature is used to characterize the installed application T i Similarity with the business object to be recommended. After the computer device obtains N attention application features between the N installed applications and the business object to be recommended, that is, one installed application corresponds to one attention application feature, the N attention application features can be spliced ​​to obtain the attention implicit features between the business object to be recommended and the N installed applications, that is, the degree of association between the business object to be recommended and the N installed applications.

[0177] Optionally, the calculation formula for obtaining the attention implicit feature between the to-be-recommended business object and the N installed applications by the computer device may be the following formula (2):

[0178]

[0179] Among them, Attnetion(Q1,K1,V1) in formula (2) refers to the implicit attention feature between the business object to be recommended and N installed applications, Q1 refers to the business feature vector corresponding to the business object to be recommended, K1 T refers to the transposed implicit features corresponding to the installed applications, d k1 It refers to the vector dimension of the transposed implicit features corresponding to each installed application, which is used to convert the first vector dot product (i.e., attention matrix) between the transposed implicit feature vector corresponding to the installed application and the business feature vector into a standard normal distribution. Softmax refers to the normalization processing, and V1 refers to the transposed implicit features corresponding to the installed application.

[0180] like Figure 5 As shown, Figure 5 is a schematic diagram of a method for obtaining implicit features of attention provided by an embodiment of the present application, such as Figure 5 As shown, after the computer device obtains the implicit feature 50b corresponding to each installed application, it can perform transposition processing on the implicit feature 50b corresponding to each installed application to obtain the transposed implicit feature vector 50c corresponding to each installed application. The computer device can perform vector product of the business feature vector 50a corresponding to the business object to be recommended and the transposed implicit feature vector 50c corresponding to each installed application to obtain the first vector dot product 50d between the business feature vector 50a corresponding to the business object to be recommended and the transposed implicit feature vector 50c corresponding to each installed application. The computer device can use the first vector dot product as an attention matrix to perform weighted processing on the implicit feature 50b corresponding to each installed application to obtain the attention implicit feature 50e between the business object to be recommended and the N installed applications. Among them, the specific implementation content of the method for obtaining the attention implicit feature can refer to the content of step S103, and the embodiments of the present application will not be repeated here.

[0181] S104: predicting the user interest of the target user in the business object to be recommended based on the user attribute information and the implicit feature of attention.

[0182] Specifically, after the computer device obtains the implicit attention feature between the business object to be recommended and N installed applications, it can extract the associated features of the user attribute information and the implicit attention feature corresponding to the target user, obtain the object interest feature between the target user and the business object to be recommended, input the object interest feature into the classifier in the target prediction target, classify the object interest feature, obtain the prediction score corresponding to the object interest feature, and determine the prediction score corresponding to the object interest feature as the user interest of the target user for the business object to be recommended. The computer device can provide business services for the target user based on the user interest, such as when the user interest is greater than the interest threshold, push the business object to be recommended to the target user, or push the business service associated with the business object to be recommended to the target user.

[0183] In an embodiment of the present application, the business object to be recommended is obtained by obtaining user feature information corresponding to the target user; the user feature information includes user attribute information corresponding to the target user, and application feature information corresponding to N installed applications associated with the target user, and feature encoding is performed on the application feature information corresponding to the N installed applications to obtain implicit features corresponding to each installed application. Through the implicit features corresponding to each installed application, the associated feature information between each installed application can be better extracted; the associated features are extracted from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and the attention implicit features between the business object to be recommended and the N installed applications are obtained. The attention implicit features are obtained by self-attention mechanism. The importance of different features can be better extracted through the self-attention mechanism, so as to focus on extracting features with greater importance, which can increase the accuracy of feature extraction for the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the attention implicit features can better characterize the implicit relationship (such as association relationship) between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended; according to the user attribute information and the attention implicit features, the user interest of the target user for the business object to be recommended is predicted, and the user interest of the target user for the business object to be recommended is predicted by using the attention implicit features and user attribute information that contain the implicit relationship between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the user interest can be predicted more accurately. In the present application, the prediction accuracy of the user interest of the target user for the business object to be recommended can be improved.

[0184] See also Figure 6 , Figure 6is a flow chart of a data processing method provided in an embodiment of the present application. The data processing method can be executed by a computer device, which can be a server (such as the above-mentioned Figure 1 Server 10 in the example above) or a user terminal (such as the one described above) Figure 1 Any user terminal in a user terminal cluster), or a system consisting of a server and a user terminal, which is not limited in this application. Figure 6 As shown, the data processing method may include steps S201-S208.

[0185] S201, obtaining user feature information corresponding to a target user, and obtaining a business object to be recommended.

[0186] S202, feature encoding is performed on application feature information corresponding to N installed applications to obtain implicit features corresponding to each installed application.

[0187] S203, extracting associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, to obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0188] For details of steps S201-S203, please refer to Figure 3 The described contents have been implemented in this application and will not be repeated here.

[0189] S204, performing vector conversion on the user attribute information to obtain a user attribute vector corresponding to the user attribute information, inputting the user attribute vector into a perception layer in a target prediction model, and outputting a global feature vector corresponding to the user attribute vector through the perception layer.

[0190] Specifically, the computer device can perform vector conversion on the user attribute information corresponding to the target object to obtain a user attribute vector corresponding to the user attribute information, wherein the vector conversion method can be Embedding (embedding network, which can convert discrete variables into continuous vectors, and the conversion into continuous vectors can not only reduce the spatial dimension of discrete variables, but also meaningfully represent discrete variables), One-hot encoding (One-Hot Encoding, also known as unique hot code, used to convert discrete variables into binary vectors), Word2vec encoding (a vector conversion method used to map each feature to a vector, which can be used to represent the relationship between features), etc. After the computer device obtains the user attribute vector, it can input the user attribute vector into the perception layer in the target prediction model, perform nonlinear combination on the user attribute vector through the perception layer, and then perform feature extraction on the user attribute vector to obtain a global feature vector corresponding to the user attribute vector.

[0191] Optionally, the specific method of the computer device outputting the global feature vector corresponding to the user attribute vector through the perception layer may include: inputting the user attribute vector into the perception layer in the target prediction model, and performing nonlinear combination on the user attribute vector based on the bias coefficient and neural matrix in the perception layer to obtain the combined feature vector corresponding to the user attribute vector. The combined feature vector is activated by the activation parameter in the perception layer to obtain the global feature vector corresponding to the user attribute vector.

[0192] Specifically, the computer device can input the user attribute vector into the perception layer in the target prediction model, obtain the neural matrix and bias coefficient in the perception layer, and perform nonlinear combination on the user attribute vector according to the neural matrix and bias coefficient to obtain a combined feature vector corresponding to the user attribute vector. After the computer device obtains the combined feature vector, it can activate the combined feature vector through the activation parameters in the perception layer to obtain a global feature vector corresponding to the user attribute vector. Among them, the number of perception layers in the target prediction model can be one or more, which can be set according to specific needs, and the embodiment of the present application is not limited here.

[0193] The computer device performs nonlinear combination on the user attribute vector according to the neural matrix and the bias coefficient, and obtains the calculation method of the combined feature vector corresponding to the user attribute vector, which can be calculated by the following formula (3):

[0194] C=qω j +b (3)

[0195] Where C in formula (3) refers to the combined feature vector corresponding to the user attribute vector, q refers to the user attribute vector, ω j refers to the neural matrix in the perception layer, and b refers to the bias coefficient in the perception layer.

[0196] The computer device activates the combined feature vector to obtain the global feature vector corresponding to the user attribute vector, which can be calculated using the following formula (4):

[0197] y=a(C) (4)

[0198] Among them, y refers to the global feature vector corresponding to the user attribute vector, a() refers to the activation parameter (i.e., activation function), and C refers to the combined feature vector corresponding to the user attribute vector.

[0199] S205, input the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature into the second self-attention network in the target prediction model, transpose the global feature vector to obtain the transposed global feature vector corresponding to the global feature vector, and obtain the second vector dot product between the transposed global feature vector and the attention implicit feature vector.

[0200] S206, normalize the second vector dot product to obtain a normalized second vector dot product.

[0201] Specifically, after the computer device obtains the global feature vector corresponding to the user attribute vector, the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature can be input into the second self-attention network in the target prediction model, the global feature vector is transposed to obtain the transposed global feature vector corresponding to the global feature vector, and the transposed global feature vector and the attention implicit feature vector are dot-producted to obtain the second vector dot product. The second vector dot product is used to characterize the similarity between the global feature vector and the attention implicit feature vector of the target user, and is used to predict the target user's interest in the business object to be recommended. After the computer device obtains the second vector dot product, the second vector dot product can be normalized to obtain the normalized second vector dot product. The purpose of normalizing the second vector dot product is to enhance the comparability between the features in the second vector dot product, so that the indicators corresponding to each feature are at the same order of magnitude, which is convenient for comprehensive comparison, and the normalized second vector dot product can also be more stable.

[0202] S207 , performing weighted processing on the global feature vector according to the dot product of the normalized second vector, to obtain an associated implicit feature between the target user and the business object to be recommended.

[0203] Specifically, the computer device obtains the dot product of the normalized second vector, and can determine the dot product of the normalized second vector as the attention matrix in the second self-attention network. The attention matrix is ​​used to perform weighted processing on the global feature vector corresponding to the target user to obtain the associated implicit features between the target user and the business object to be recommended.

[0204] The calculation method of the associated implicit feature between the target user of the computer device and the business object to be recommended can be the following formula (5):

[0205]

[0206] Among them, Attnetion(Q2,K2,V2) in formula (5) refers to the implicit feature of the association between the target user and the business object to be recommended, softmax() refers to the normalization function, Q2 refers to the implicit feature of attention between the business object to be recommended and N installed applications, K2 T refers to the global feature vector after transposition processing, V2 refers to the global feature vector corresponding to the user attribute feature, and d k2 is the vector dimension.

[0207] S208 , predicting the user interest of the target user in the business object to be recommended based on the associated implicit features between the target user and the business object to be recommended.

[0208] Specifically, after the associated implicit features between the target user of the computer device and the business object to be recommended are obtained, the associated implicit features can be input into the classifier in the target prediction model, and the associated implicit features can be classified by the classifier to obtain the classification results corresponding to the associated implicit features, and the user interest of the target user for the business object to be recommended is determined according to the classification results. For example, the prediction score or prediction probability corresponding to the associated implicit features is obtained by the classifier, and the prediction score or prediction probability is determined as the user interest of the target user for the business object to be recommended. After the computer device obtains the user interest, it can provide business services for the target user according to the user interest.

[0209] Optionally, after the computer device obtains the target user's interest in the business object to be recommended, the specific method of providing business services to the target user based on the user interest may include: if the user interest is greater than or equal to the interest threshold, the object to be recommended is determined as the target recommendation object, and the target recommendation object is sent to the target user terminal corresponding to the target user, so that the target user terminal displays the target recommendation object.

[0210] Specifically, after the computer device obtains the user's interest, it determines whether the user's interest is greater than the interest threshold. If the user's interest is greater than or equal to the interest threshold, it can be determined that the target user is interested in the recommended business object, and the recommended business object can be determined as the target recommended object, and the target user terminal corresponding to the target user sends the target recommended object, so that the target user terminal displays the target recommended object, thereby promoting the recommended business object. If the user's interest is less than the interest threshold, it can be determined that the target user is not interested in the recommended business object, and the recommended business object will not be pushed to the target user, which can avoid interference with the target user and achieve accurate recommendation. The recommended business object can also be determined as a temporarily invalid object, which means that within a certain time threshold, the computer device no longer predicts its user interest, which can reduce the workload of the computer device and the amount of calculation, and improve the prediction efficiency of the target recommended object that the target user is interested in.

[0211] like Figure 7 As shown, Figure 7 is a schematic diagram of a method for obtaining user interest provided by an embodiment of the present application, such as Figure 7As shown, after the computer device obtains the user attribute information 70a corresponding to the target user, the application feature information 70b corresponding to each installed application, and the business feature information 70c corresponding to the business object to be recommended, the user attribute information can be input into the embedding layer 70d, and the user attribute information can be vector-converted to obtain the user attribute vector 70g corresponding to the user attribute information. The computer device can input the application feature information corresponding to each installed application into the embedding layer 70e, and the application feature information corresponding to each installed application can be vector-converted to obtain the application feature vector 70h corresponding to each installed application. The computer device can input the business feature information corresponding to the business object to be recommended into the embedding layer 70f, and the business feature information can be vector-converted to obtain the business feature vector 70k corresponding to the business object to be recommended. After the computer device obtains the user attribute vector 70g, the application feature vector 70h and the business feature vector 70k, the user attribute vector 70g, the application feature vector 70h and the business feature vector 70k can be input into the target prediction model 70r, and the global feature vector 70n corresponding to the user attribute vector can be obtained through the perception layer 70i in the target prediction model 790r. For details, please refer to Figure 6 The contents of steps S205-S207 in the target prediction model are not repeated here. The feature encoding layer 70j in the target prediction model performs feature encoding on the application feature vector 70h corresponding to each installed application to obtain the implicit features 70l corresponding to each installed application, such as the implicit features a1 corresponding to installed application 1, the implicit features a2 corresponding to installed application 2, ..., the implicit features an corresponding to installed application n. For specific implementation content, please refer to Figure 3 After obtaining the implicit features 701 corresponding to each installed application, the first self-attention network 70m in the target prediction model 70r can be used to extract the associated features of the implicit features 701 corresponding to each installed application and the business feature vector 70k corresponding to the business object to be recommended, and obtain the attention implicit features 70o between the business object to be recommended and the N installed applications. For specific implementation content, please refer to Figure 3 The contents of step S103 in the present application embodiment will not be repeated here. The computer device can input the global feature vector 70h and the attention implicit feature 70o corresponding to the target user into the second self-attention network 70p in the target prediction model 70r, and output the associated implicit features between the target user and the business object to be recommended through the second self-attention network 70p. For details, please refer to Figure 6After obtaining the associated implicit features between the target user and the business object to be recommended, the associated implicit features between the target user and the business object to be recommended can be classified by the classifier 70q in the target prediction model 70r to obtain the user interest 70s of the target user for the business object to be recommended.

[0212] like Figure 8 As shown, Figure 8 This is an application scenario diagram for recommending business objects based on user interest provided by an embodiment of the present application, such as Figure 8 As shown, the business object to be recommended may refer to game information. When the business personnel 80b completes the game information about the game, the target prediction model can be used to predict the users who are interested in the game information, and the game information can be recommended to the interested users, thereby achieving accurate push and avoiding unnecessary interference to the users. Figure 8 As shown, after the business personnel 80b has completed the production of the game information A, the game information A can be pushed through the user terminal 80a, such as clicking the confirmation button for pushing the game information A in the display page 80c of the user terminal 80a. The user terminal 80a receives the trigger operation of the business personnel 80b, displays the display interface 80d, and obtains the user set that can be recommended at present, which includes one or more users, and obtains the user i from the user set, where i is a positive integer less than or equal to the number of users in the user set, such as i can be 1, 2, 3... The user terminal 80a can obtain the user attribute information corresponding to the user i, and obtain the application feature information corresponding to the N game applications installed by the user i, and input the user attribute information corresponding to the user i, the application feature information corresponding to the N installed game applications, and the business feature information corresponding to the game information A into the target prediction model 80e, and output the user interest of each user in the user set for the game information A through the target prediction model. The specific content of the target prediction model predicting the user interest can be found in Figure 3 The contents of steps S101-S104 or Figure 6The contents of steps S201-S208 in the embodiment of the present application will not be repeated here. After the user terminal 80a obtains the user interest of each user for the game information A, it can display the user interest of each user for the game information to the business personnel 80b in the display interface 80f, such as user 1's user interest for game information A is 90, user 2's user interest for game information A is 82, etc. A push button can be set after each user interest, and when the push button is clicked, the game information A is pushed to the corresponding user. The business personnel can set it to push directly to the user whose user interest is higher than the interest threshold when the user interest is higher than the interest threshold. For example, when the business personnel 80b clicks the push button corresponding to the user interest of user 1 in the display interface 80g of the user terminal 80a, the user terminal 80a can push the game information A to the user terminal 80h corresponding to user 1 (i.e. user 80i). After receiving the game information, the user terminal 80h can display the game information A on the display page 80j of the user terminal 80h, thereby achieving accurate promotion of the game information A and avoiding unnecessary interference to the user.

[0213] In an embodiment of the present application, the business object to be recommended is obtained by obtaining user feature information corresponding to the target user; the user feature information includes user attribute information corresponding to the target user and application feature information corresponding to N installed applications associated with the target user, and feature encoding is performed on the application feature information corresponding to the N installed applications to obtain implicit features corresponding to each installed application. Through the implicit features corresponding to each installed application, the associated feature information between each installed application can be better extracted; the associated features are extracted from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and the attention implicit features between the business object to be recommended and the N installed applications are obtained. The attention implicit features are obtained by the self-attention mechanism for each The implicit features corresponding to the installed applications and the business feature information corresponding to the business object to be recommended are extracted by correlating the features. The importance of different features can be better extracted through the self-attention mechanism, so as to focus on extracting features with greater importance, which can increase the accuracy of feature extraction for the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the attention implicit features can better characterize the implicit relationship (such as correlation relationship) between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended; through the self-attention mechanism, the user attribute information and the attention implicit features are extracted by correlating the features, and the user interest of the target user for the business object to be recommended is predicted, so that the user interest can be predicted more accurately. In the present application, the prediction accuracy of the user interest of the target user for the business object to be recommended can be improved.

[0214] See also Fig. 9 , Fig. 9 is a flow chart of a data processing method provided in an embodiment of the present application. The data processing method can be executed by a computer device, which can be a server (such as the above-mentioned Figure 1 Server 10 in the example above) or a user terminal (such as the one described above) Figure 1 Any user terminal in a user terminal cluster), or a system consisting of a server and a user terminal, which is not limited in this application. Fig. 9 As shown, the data processing method may include steps S301-S306.

[0215] S301, obtain the initial prediction model, sample user feature information corresponding to the sample user, sample business objects, and the sample user's interest label for the sample business objects; the sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer.

[0216] Specifically, the computer device may train the initial prediction model to obtain a target prediction model for predicting the user's interest in the business object. The computer device may obtain the initial prediction model, sample user feature information corresponding to the sample user, the sample business object, and the sample user's interest label for the sample business object, and the interest label may be obtained by manual annotation. The sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, respectively, and the sample business object and the M sample installed applications have the same attribute type.

[0217] S302, using the initial prediction model, feature encoding the sample application feature information corresponding to each sample installed application, to obtain the sample implicit feature corresponding to each sample installed application.

[0218] Specifically, the computer device can input sample user feature information and sample business objects corresponding to the sample users into the initial prediction model, and use the initial prediction model to feature encode the sample application feature information corresponding to each sample installed application to obtain sample implicit features corresponding to each sample installed application.

[0219] S303, extracting associated features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object, to obtain the sample attention implicit features between the sample business object and the sample installed application.

[0220] S304: predicting the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit feature.

[0221] Specifically, the computer device may perform associated feature extraction on the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object in the initial prediction model to obtain the sample attention implicit features between the sample business object and the sample installed application. After obtaining the sample attention implicit features, the computer device may predict the predicted user interest of the sample user for the sample business object in the initial prediction model based on the sample user attribute information and the sample attention implicit features.

[0222] The specific implementation of steps S301-S306 can be found in Figure 3 or Figure 6 The described contents will not be repeated here in the embodiments of the present application.

[0223] S305: Determine a loss function corresponding to the initial prediction model according to the predicted user interest and the interest label.

[0224] S306, adjusting the parameters of the initial prediction model according to the loss function, and when the initial prediction model after the parameter adjustment meets the convergence condition, determining the target prediction model from the initial prediction model that meets the convergence condition; the target prediction model is used to predict the user interest in the business object.

[0225] Specifically, after the computer device obtains the predicted user interest corresponding to the initial prediction model, it can determine the loss function between the predicted user interest and the interest label, and determine the loss value (i.e., error) between the predicted user interest and the interest label according to the loss function. According to the loss value between the predicted user interest and the interest label, the adjustment model parameters used to adjust the parameters of the initial prediction model are determined, and the initial model parameters in the initial prediction model are updated according to the adjustment model parameters to obtain the initial prediction model after the parameter update. If the initial prediction model after the parameter update meets the convergence condition, the initial prediction model that meets the convergence condition can be determined as the target prediction model, and the target prediction model is used to predict the user interest of the target user for the business object. If the initial prediction model after the parameter update does not meet the convergence condition, the initial prediction model continues to be trained until the initial prediction model reaches the convergence condition. The convergence condition may refer to that the loss value corresponding to the initial prediction model is less than or equal to the loss threshold, or that the number of training times corresponding to the initial prediction model is the target number of training times, etc.

[0226] In an embodiment of the present application, by obtaining an initial prediction model, sample user feature information corresponding to sample users, sample business objects, and sample user interest labels for sample business objects, the initial prediction model is used to perform feature encoding on the sample application feature information corresponding to each sample installed application, and obtain sample implicit features corresponding to each sample installed application; perform associated feature extraction on the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object, and obtain sample attention implicit features between the sample business object and the sample installed application; predict the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit features, and determine the loss function corresponding to the initial prediction model based on the predicted user interest and the interest label; adjust the parameters of the initial prediction model based on the loss function, and when the initial prediction model after parameter adjustment meets the convergence condition, determine the target prediction model with the initial prediction model that meets the convergence condition. Through this application, the prediction accuracy of the target prediction model for user interest prediction can be improved.

[0227] See also Fig.10 , Fig.10 1 is a schematic diagram of the structure of a data processing device 1 provided in an embodiment of the present application. The above-mentioned data processing device 1 may be a computer program (including program code) running in a computer device, for example, the data processing device 1 is an application software; the data processing device 1 may be used to execute the corresponding steps in the data processing method provided in an embodiment of the present application. Fig.10 As shown, the data processing device 1 may include: a first acquisition module 11, a first feature encoding module 12, a first associated feature extraction module 13, a first prediction module 14, a first determination module 15, and a sending module 16.

[0228] The first acquisition module 11 is used to acquire user characteristic information corresponding to the target user and acquire the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0229] A first feature encoding module 12 is used to perform feature encoding on application feature information corresponding to N installed applications to obtain implicit features corresponding to each installed application;

[0230] A first correlation feature extraction module 13 is used to extract correlation features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0231] The first prediction module 14 is used to predict the user interest of the target user for the business object to be recommended based on the user attribute information and the implicit feature of attention.

[0232] Wherein, the first feature encoding module 12 includes:

[0233] The first vector conversion unit 1201 is used to perform vector conversion on the application feature information corresponding to each of the N installed applications to obtain an application feature vector corresponding to each installed application;

[0234] The feature encoding unit 1202 is used to input the application feature vector corresponding to each installed application into the target prediction model, and based on the feature encoding network in the target prediction model, feature encode the application feature vector corresponding to each installed application to obtain the implicit features corresponding to each installed application; the target prediction model is used to predict the user's interest in the business object.

[0235] Among them, the feature encoding network includes a self-attention layer and a forward feedback layer;

[0236] The feature encoding unit 1202 includes:

[0237] A correlation feature extraction subunit is used to extract correlation features from the application feature vectors corresponding to each installed application based on the attention matrix in the self-attention layer, so as to obtain a target self-attention vector corresponding to each installed application;

[0238] The first nonlinear combination subunit is used to perform nonlinear combination on the target self-attention vectors corresponding to each installed application based on the forward feedback layer to obtain the implicit features corresponding to each installed application.

[0239] Among them, the self-attention layer includes a self-attention sublayer, a first residual connection sublayer and a first normalization sublayer;

[0240] The associated feature extraction subunit is specifically used for:

[0241] Based on the attention matrix in the self-attention sublayer, the application feature vectors corresponding to each installed application are weighted to obtain the initial self-attention vector corresponding to each application feature vector;

[0242] Based on the first residual connection sublayer, residual connection is performed on the initial self-attention vectors corresponding to each application feature vector, so as to obtain the first residual feature vectors corresponding to each application feature vector;

[0243] Based on the first normalization sublayer, the first residual feature vector corresponding to each application feature vector is normalized to obtain the target self-attention vector corresponding to each installed application.

[0244] The associated feature extraction subunit is also specifically used for:

[0245] Obtain the first matrix, the second matrix, and the third matrix in the self-attention sublayer, and perform weighted processing on the application feature vectors corresponding to each installed application according to the first matrix, the second matrix, and the third matrix, respectively, to obtain the query vector, the key vector, and the value vector corresponding to each installed application;

[0246] For installed applications i The corresponding key vector is transposed to obtain the installed application T i Corresponding transposed key vector, get installed application T i The corresponding query vector and installed applications T i Apply vector dot product between corresponding transposed key vectors; installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0247] According to the dot product of the application vector, the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

[0248] The forward feedback layer includes a fully connected sublayer, a second residual connection sublayer, and a second normalization sublayer;

[0249] The first nonlinear combination subunit is specifically used for:

[0250] Based on the fully connected sublayer, convolution is performed on the target self-attention vector corresponding to each installed application to obtain the convolution feature vector corresponding to each installed application.

[0251] Based on the second residual connection sublayer, performing residual connection on the convolution feature vectors corresponding to each installed application, to obtain a second residual feature vector corresponding to each convolution feature vector;

[0252] Based on the second normalization sublayer, the second residual feature vector corresponding to each convolution feature vector is normalized to obtain an implicit feature corresponding to each installed application.

[0253] Wherein, the first correlation feature extraction module 13 includes:

[0254] The second vector conversion unit 1301 is used to perform vector conversion on the service feature information corresponding to the service object to be recommended, so as to obtain a service feature vector corresponding to the service object to be recommended;

[0255] The transposition processing unit 1302 is used to input the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, and perform transposition processing on the implicit features corresponding to each installed application to obtain the transposed implicit feature vector corresponding to each installed application; the target prediction model is used to predict the user interest in the business object;

[0256] The associated feature extraction unit 1303 is used to extract associated features from the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended based on the first self-attention network, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications.

[0257] Wherein, the associated feature extraction unit 1303 includes:

[0258] Get subunit, used to get installed applications T based on the first self-attention network i The first vector dot product between the corresponding transposed implicit feature vector and the business feature vector; Installed application T i Belongs to N installed applications, i is a positive integer less than or equal to N;

[0259] A normalization subunit, used for normalizing the first vector dot product to obtain a normalized first vector dot product;

[0260] The weighted processing subunit is used to apply T to the installed i The corresponding implicit features are weighted to obtain the installed applications T i Attention application features between the business object to be recommended;

[0261] Determine the subunit for installing the application T i Attention application features between the business object to be recommended and the N installed applications are determined, and attention implicit features between the business object to be recommended and the N installed applications are determined.

[0262] The first prediction module 14 includes:

[0263] The output unit 1401 is used to perform vector conversion on the user attribute information to obtain a user attribute vector corresponding to the user attribute information, input the user attribute vector into the perception layer in the target prediction model, and output a global feature vector corresponding to the user attribute vector through the perception layer; the target prediction model is used to predict the user interest in the business object;

[0264] An acquisition unit 1402 is used to input the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature into a second self-attention network in the target prediction model, transpose the global feature vector to obtain a transposed global feature vector corresponding to the global feature vector, and obtain a second vector dot product between the transposed global feature vector and the attention implicit feature vector;

[0265] A normalization processing unit 1403 is used to perform normalization processing on the second vector dot product to obtain a normalized second vector dot product;

[0266] The weighted processing unit 1404 is used to perform weighted processing on the global feature vector according to the dot product of the normalized second vector to obtain the associated implicit feature between the target user and the service object to be recommended;

[0267] The prediction unit 1405 is used to predict the user interest of the target user in the business object to be recommended according to the associated implicit features between the target user and the business object to be recommended.

[0268] The output unit 1401 includes:

[0269] The second nonlinear combination subunit is used to input the user attribute vector into the perception layer in the target prediction model, and perform nonlinear combination on the user attribute vector based on the bias coefficient and the neural matrix in the perception layer to obtain a combined feature vector corresponding to the user attribute vector;

[0270] The activation processing subunit is used to activate the combined feature vector through the activation parameters in the perception layer to obtain the global feature vector corresponding to the user attribute vector.

[0271] The data processing device 1 further includes:

[0272] A first determination module 15, configured to determine the object to be recommended as a target recommendation object if the user's interest level is greater than or equal to an interest level threshold;

[0273] The sending module 16 is used to send the target recommended object to the target user terminal corresponding to the target user, so that the target user terminal displays the target recommended object.

[0274] According to one embodiment of the present application, Fig.10The various modules in the data processing device 1 shown can be separately or all combined into one or several units to constitute, or one (some) of the units can be further divided into multiple functionally smaller sub-units, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above modules are divided based on logical functions. In practical applications, the functions of one module can also be implemented by multiple units, or the functions of multiple modules can be implemented by one unit. In other embodiments of the present application, the data processing device 1 may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0275] In an embodiment of the present application, the business object to be recommended is obtained by obtaining user feature information corresponding to the target user; the user feature information includes user attribute information corresponding to the target user and application feature information corresponding to N installed applications associated with the target user, and feature encoding is performed on the application feature information corresponding to the N installed applications to obtain implicit features corresponding to each installed application. Through the implicit features corresponding to each installed application, the associated feature information between each installed application can be better extracted; the associated features are extracted from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and the attention implicit features between the business object to be recommended and the N installed applications are obtained. The attention implicit features are obtained by the self-attention mechanism for each The implicit features corresponding to the installed applications and the business feature information corresponding to the business object to be recommended are extracted by correlating the features. The importance of different features can be better extracted through the self-attention mechanism, so as to focus on extracting features with greater importance, which can increase the accuracy of feature extraction for the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, so that the attention implicit features can better characterize the implicit relationship (such as correlation relationship) between the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended; through the self-attention mechanism, the user attribute information and the attention implicit features are extracted by correlating the features, and the user interest of the target user for the business object to be recommended is predicted, so that the user interest can be predicted more accurately. In the present application, the prediction accuracy of the user interest of the target user for the business object to be recommended can be improved.

[0276] See also Fig.11 , Fig.11 : is a schematic diagram of the structure of a data processing device 2 provided in an embodiment of the present application. The above-mentioned data processing device 2 may be a computer program (including program code) running in a computer device, for example, the data processing device 2 is an application software; the data processing device 2 may be used to execute the corresponding steps in the data processing method provided in an embodiment of the present application. Fig.11 As shown, the data processing device 2 may include: a second acquisition module 21, a second feature encoding module 22, a second associated feature extraction module 23, a second prediction module 24, a second determination module 25, and a third determination module 26.

[0277] The second acquisition module 21 is used to acquire the initial prediction model, sample user feature information corresponding to the sample user, the sample business object, and the sample user's interest label for the sample business object; the sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, respectively; the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer;

[0278] The second feature encoding module 22 is used to use the initial prediction model to perform feature encoding on the sample application feature information corresponding to each sample installed application, so as to obtain the sample implicit feature corresponding to each sample installed application;

[0279] The second correlation feature extraction module 23 is used to extract correlation features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object, so as to obtain the sample attention implicit features between the sample business object and the sample installed application;

[0280] A second prediction module 24, configured to predict the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit feature;

[0281] A second determination module 25, for determining a loss function corresponding to the initial prediction model according to the predicted user interest and the interest label;

[0282] The third determination module 26 is used to adjust the parameters of the initial prediction model according to the loss function. When the initial prediction model after parameter adjustment meets the convergence condition, the target prediction model is determined by the initial prediction model that meets the convergence condition; the target prediction model is used to predict the user interest in the business object.

[0283] According to one embodiment of the present application, Fig.11The various modules in the data processing device 2 shown can be separately or all combined into one or several units to constitute, or one (some) of the units can be further divided into multiple smaller sub-units in function, and the same operation can be achieved without affecting the realization of the technical effects of the embodiments of the present application. The above modules are divided based on logical functions. In practical applications, the functions of a module can also be implemented by multiple units, or the functions of multiple modules can be implemented by one unit. In other embodiments of the present application, the data processing device 2 may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0284] In an embodiment of the present application, by obtaining an initial prediction model, sample user feature information corresponding to sample users, sample business objects, and sample user interest labels for sample business objects, the initial prediction model is used to perform feature encoding on the sample application feature information corresponding to each sample installed application, and obtain sample implicit features corresponding to each sample installed application; perform associated feature extraction on the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object, and obtain sample attention implicit features between the sample business object and the sample installed application; predict the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit features, and determine the loss function corresponding to the initial prediction model based on the predicted user interest and the interest label; adjust the parameters of the initial prediction model based on the loss function, and when the initial prediction model after parameter adjustment meets the convergence condition, determine the target prediction model with the initial prediction model that meets the convergence condition. Through this application, the prediction accuracy of the target prediction model for user interest prediction can be improved.

[0285] See also Fig.12 , Fig.12 Schematic diagram of the structure of a computer device provided in an embodiment of the present application. Fig.12As shown, the above-mentioned computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the above-mentioned computer device 1000 may also include: a target user interface 1003, and at least one communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. Among them, the target user interface 1003 may include a display screen (Display), a keyboard (Keyboard), and the target user interface 1003 may optionally include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 1005 may optionally also be at least one storage device located away from the aforementioned processor 1001. As Fig.12 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a target user interface module, and a device control application program.

[0286] exist Fig.12 In the computer device 1000 shown, the network interface 1004 can provide a network communication function; the target user interface 1003 is mainly used to provide an input interface for the target user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve:

[0287] Obtain user characteristic information corresponding to the target user and obtain the business object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, wherein the N installed applications have the same attribute type as the business object to be recommended, and N is a positive integer;

[0288] Feature encoding is performed on the application feature information corresponding to each of the N installed applications to obtain an implicit feature corresponding to each of the installed applications;

[0289] Extract associated features from the implicit features corresponding to each installed application and the business feature information corresponding to the business object to be recommended, and obtain the attention implicit features between the business object to be recommended and the N installed applications;

[0290] According to user attribute information and implicit attention features, the user interest of the target user in the business object to be recommended is predicted.

[0291] It should be understood that the computer device 1000 described in the embodiment of the present application can execute the above Figure 3 or Figure 6The description of the data processing method in the corresponding embodiment can also be performed as described above. Figure 3 The description of the data processing device 1 in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated here either.

[0292] See also Fig.13 , Fig.13 Schematic diagram of the structure of a computer device provided in an embodiment of the present application. Fig.13 As shown, the computer device 2000 may include: a processor 2001, a network interface 2004 and a memory 2005. In addition, the above-mentioned computer device 2000 may also include: a user interface 2003, and at least one communication bus 2002. Among them, the communication bus 2002 is used to realize the connection and communication between these components. Among them, the user interface 2003 may include a display screen (Display), a keyboard (Keyboard), and the optional user interface 2003 may also include a standard wired interface and a wireless interface. Optionally, the network interface 2004 may include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 2005 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. Optionally, the memory 2005 may also be at least one storage device located away from the aforementioned processor 2001. As Fig.13 As shown, the memory 2005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application program.

[0293] In such Fig.13 In the computer device 2000 shown, the network interface 2004 can provide a network communication function; the user interface 2003 is mainly used to provide an input interface for the user; and the processor 2001 can be used to call the device control application stored in the memory 2005 to achieve:

[0294] Obtain an initial prediction model, sample user feature information corresponding to the sample user, a sample business object, and a sample user's interest label for the sample business object; the sample user feature information includes sample user attribute information corresponding to the sample user, and application feature information corresponding to M sample installed applications associated with the sample user, respectively; the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer;

[0295] Using the initial prediction model, feature encoding is performed on the sample application feature information corresponding to each sample installed application, so as to obtain the sample implicit features corresponding to each sample installed application;

[0296] Extracting associated features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object to obtain the sample attention implicit features between the sample business object and the sample installed application;

[0297] Predicting the predicted user interest of the sample user for the sample business object based on the sample user attribute information and the sample attention implicit features;

[0298] Determine the loss function corresponding to the initial prediction model based on the predicted user interest and interest label;

[0299] According to the loss function, the parameters of the initial prediction model are adjusted. When the initial prediction model after the parameter adjustment meets the convergence condition, the initial prediction model that meets the convergence condition is used to determine the target prediction model; the target prediction model is used to predict the user interest in the business object.

[0300] It should be understood that the computer device 2000 described in the embodiment of the present application can execute the above Fig. 9 The description of the data processing method in the corresponding embodiment can also be performed as described above. Fig. 9 The description of the data processing device 2 in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated here either.

[0301] In addition, it should be pointed out that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the data processing device 1 and the data processing device 2 mentioned above, and the computer program includes program instructions. When the processor executes the program instructions, it can execute the above-mentioned Figure 3 , Figure 6 as well as Fig. 9 The description of the data processing method in the corresponding embodiment will therefore not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application. As an example, program instructions may be deployed on a computing device for execution, or on multiple computing devices located at one location, or on multiple computing devices distributed at multiple locations and interconnected by a communication network. Multiple computing devices distributed at multiple locations and interconnected by a communication network can constitute a blockchain system.

[0302] In addition, it should be noted that: the embodiment of the present application also provides a computer program product or a computer program, which may include computer instructions, and the computer instructions may be 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 may execute the computer instructions, so that the computer device executes the above Figure 3 , Figure 6 or Fig. 9 The description of the data processing method in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated. For technical details not disclosed in the computer program product or computer program embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0303] It should be noted that, for the above-mentioned various method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0304] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs.

[0305] The modules in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0306] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The above-mentioned storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0307] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A data processing method, characterized in that: include: Obtain user characteristic information corresponding to the target user, and obtain the service object to be recommended; the user characteristic information includes user attribute information corresponding to the target user, and application characteristic information corresponding to N installed applications associated with the target user, respectively, wherein the N installed applications and the service object to be recommended have the same attribute type, and N is a positive integer; Performing feature encoding on the application feature information corresponding to the N installed applications respectively, to obtain implicit features corresponding to each installed application respectively; Extracting associated features from the implicit features corresponding to each of the installed applications and the business feature information corresponding to the business object to be recommended, to obtain implicit features of attention between the business object to be recommended and the N installed applications; The user interest of the target user in the business object to be recommended is predicted according to the user attribute information and the attention implicit feature.

2. The method according to claim 1 is characterized in that the feature encoding of the application feature information corresponding to the N installed applications to obtain the implicit features corresponding to each installed application comprises: Performing vector conversion on the application feature information corresponding to each of the N installed applications to obtain an application feature vector corresponding to each of the installed applications; The application feature vector corresponding to each installed application is input into the target prediction model, and based on the feature encoding network in the target prediction model, the application feature vector corresponding to each installed application is feature encoded to obtain the implicit features corresponding to each installed application; the target prediction model is used to predict the user's interest in the business object.

3. The method according to claim 2, characterized in that the feature encoding network includes a self-attention layer and a forward feedback layer; The feature encoding network in the target prediction model is based on which the application feature vectors corresponding to each installed application are feature encoded to obtain the implicit features corresponding to each installed application, including: Based on the attention matrix in the self-attention layer, extract associated features from the application feature vectors corresponding to each installed application to obtain a target self-attention vector corresponding to each installed application; Based on the forward feedback layer, a nonlinear combination is performed on the target self-attention vectors corresponding to each installed application to obtain an implicit feature corresponding to each installed application.

4. The method according to claim 3, characterized in that the self-attention layer includes a self-attention sublayer, a first residual connection sublayer and a first normalization sublayer; The extracting associated features of the application feature vectors corresponding to each installed application based on the attention matrix in the self-attention layer to obtain the target self-attention vectors corresponding to each installed application includes: Based on the attention matrix in the self-attention sublayer, weighted processing is performed on the application feature vectors corresponding to each installed application to obtain an initial self-attention vector corresponding to each application feature vector; Based on the first residual connection sublayer, performing residual connection on the initial self-attention vectors corresponding to each of the application feature vectors, to obtain first residual feature vectors corresponding to each of the application feature vectors; Based on the first normalization sublayer, the first residual feature vector corresponding to each application feature vector is normalized to obtain a target self-attention vector corresponding to each installed application.

5. The method according to claim 4, characterized in that, based on the attention matrix in the self-attention sublayer, weighted processing is performed on the application feature vectors corresponding to each installed application to obtain the initial self-attention vectors corresponding to each application feature vector, including: Obtaining a first matrix, a second matrix, and a third matrix in the self-attention sublayer, and performing weighted processing on the application feature vectors corresponding to each installed application according to the first matrix, the second matrix, and the third matrix, respectively, to obtain a query vector, a key vector, and a value vector corresponding to each installed application; For installed applications i The corresponding key vector is transposed to obtain the installed application T i The corresponding transposed key vector is used to obtain the installed application T i The corresponding query vector and the installed application T i The application vector dot product between the corresponding transposed key vectors; the installed application T i belongs to the N installed applications, i is a positive integer less than or equal to N; According to the dot product of the application vector, the installed application T i The corresponding value vector is weighted to obtain the installed application T i The corresponding initial self-attention vector.

6. The method according to claim 3, characterized in that the forward feedback layer includes a fully connected sublayer, a second residual connection sublayer and a second normalization sublayer; The nonlinear combination of the target self-attention vectors corresponding to each installed application is performed based on the forward feedback layer to obtain the implicit features corresponding to each installed application, including: Based on the fully connected sublayer, convolution processing is performed on the target self-attention vector corresponding to each installed application to obtain a convolution feature vector corresponding to each installed application; Based on the second residual connection sublayer, performing residual connection on the convolution feature vectors corresponding to each installed application, to obtain a second residual feature vector corresponding to each convolution feature vector; Based on the second normalization sublayer, the second residual feature vector corresponding to each convolution feature vector is normalized to obtain the implicit feature corresponding to each installed application.

7. The method according to claim 1 is characterized in that the step of extracting associated features from the implicit features corresponding to each of the installed applications and the business feature information corresponding to the business object to be recommended to obtain the attention implicit features between the business object to be recommended and the N installed applications comprises: Performing vector conversion on the service feature information corresponding to the service object to be recommended to obtain a service feature vector corresponding to the service object to be recommended; Inputting the implicit features corresponding to each installed application into the first self-attention network in the target prediction model, transposing the implicit features corresponding to each installed application, and obtaining the transposed implicit feature vectors corresponding to each installed application; the target prediction model is used to predict the user interest in the business object; Based on the first self-attention network, associated features are extracted from the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications.

8. The method according to claim 7, characterized in that, based on the first self-attention network, performing correlation feature extraction on the transposed implicit feature vector corresponding to each installed application and the business feature vector corresponding to the business object to be recommended, to obtain the attention implicit feature between the business object to be recommended and the N installed applications, comprises: Based on the first self-attention network, obtain the installed applications T i The first vector dot product between the corresponding transposed implicit feature vector and the service feature vector; the installed application T i belongs to the N installed applications, i is a positive integer less than or equal to N; Normalizing the first vector dot product to obtain a normalized first vector dot product; According to the normalized first vector dot product, the installed application T i The corresponding implicit features are weighted to obtain the installed application T i Attention application characteristics between the business object to be recommended; According to the installed application T i The attention application features between the service object to be recommended and the N installed applications are determined to determine the attention implicit features between the service object to be recommended and the N installed applications.

9. The method according to claim 1, characterized in that predicting the user interest of the target user for the service object to be recommended based on the user attribute information and the attention implicit feature comprises: Performing vector conversion on the user attribute information to obtain a user attribute vector corresponding to the user attribute information, inputting the user attribute vector into a perception layer in a target prediction model, and outputting a global feature vector corresponding to the user attribute vector through the perception layer; The target prediction model is used to predict the user's interest in the business object; Inputting the global feature vector and the attention implicit feature vector corresponding to the attention implicit feature into the second self-attention network in the target prediction model, transposing the global feature vector to obtain a transposed global feature vector corresponding to the global feature vector, and obtaining a second vector dot product between the transposed global feature vector and the attention implicit feature vector; Normalizing the second vector dot product to obtain a normalized second vector dot product; According to the dot product of the normalized second vector, weighting the global feature vector is performed to obtain an associated implicit feature between the target user and the service object to be recommended; The user interest of the target user in the business object to be recommended is predicted according to the associated implicit features between the target user and the business object to be recommended.

10. The method according to claim 9, characterized in that the step of inputting the user attribute vector into a perception layer in a target prediction model and outputting a global feature vector corresponding to the user attribute vector through the perception layer comprises: Inputting the user attribute vector into the perception layer in the target prediction model, and performing nonlinear combination on the user attribute vector based on the bias coefficient and the neural matrix in the perception layer to obtain a combined feature vector corresponding to the user attribute vector; The combined feature vector is activated by using the activation parameters in the perception layer to obtain a global feature vector corresponding to the user attribute vector.

11. The method according to claim 1, characterized in that the method further comprises: If the user interest level is greater than or equal to the interest level threshold, the object to be recommended is determined as a target recommended object; The target recommended object is sent to a target user terminal corresponding to the target user, so that the target user terminal displays the target recommended object.

12. A data processing method, comprising: Acquire an initial prediction model, sample user feature information corresponding to the sample user, a sample business object, and an interest label of the sample user for the sample business object; The sample user characteristic information includes sample user attribute information corresponding to the sample user and application characteristic information corresponding to M sample installed applications associated with the sample user, the sample business object and the M sample installed applications have the same attribute type, and M is a positive integer; Using the initial prediction model, feature encoding is performed on sample application feature information corresponding to each sample installed application to obtain sample implicit features corresponding to each sample installed application; Extracting associated features from the sample implicit features corresponding to each sample installed application and the sample business feature information corresponding to the sample business object to obtain the sample attention implicit features between the sample business object and the sample installed application; Predicting the predicted user interest of the sample user for the sample business object according to the sample user attribute information and the sample attention implicit feature; Determining a loss function corresponding to the initial prediction model according to the predicted user interest and the interest label; According to the loss function, the parameters of the initial prediction model are adjusted. When the initial prediction model after the parameter adjustment meets the convergence condition, the initial prediction model that meets the convergence condition is used to determine the target prediction model; the target prediction model is used to predict the user interest in the business object.

13. A data processing device, characterized in that: include: A first acquisition module is used to acquire user characteristic information corresponding to a target user and acquire a service object to be recommended; the user characteristic information includes user attribute information corresponding to the target user and application characteristic information corresponding to N installed applications associated with the target user, wherein the N installed applications have the same attribute type as the service object to be recommended, and N is a positive integer; A feature encoding module, used to perform feature encoding on the application feature information corresponding to the N installed applications respectively, to obtain an implicit feature corresponding to each installed application respectively; A correlation feature extraction module, used to extract correlation features from the implicit features corresponding to each of the installed applications and the business feature information corresponding to the business object to be recommended, so as to obtain the attention implicit features between the business object to be recommended and the N installed applications; A prediction module is used to predict the user interest of the target user in the business object to be recommended based on the user attribute information and the attention implicit feature.

14. A computer device, characterized in that: include: Processor and memory; The memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that: A computer program is stored in the computer-readable storage medium, and the computer program is suitable for being loaded by a processor and executing the method of any one of claims 1 to 12.

Images