Object recommendation method and device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides an object recommendation method and device, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence. The method comprises the steps that object operation information of a current user and multiple pieces of object information of multiple objects are obtained, the object operation information is operation information of the current user on multiple historical objects in a preset time period, and the multiple objects comprise the multiple historical objects; determining an embedded feature vector of the current user according to the object operation information and the multiple pieces of object information; splicing the embedded feature vector and a user feature vector of the current user to obtain a target feature vector of the current user; and performing data processing on the target feature vector through a preset recommendation model to obtain a recommendation object corresponding to the current user. According to the method, the accuracy of the recommended object is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an object recommendation method, device, equipment, storage medium and program product. Background Art

[0002] In order to improve user experience, the recommendation system can recommend objects to users based on the association between users and objects.

[0003] In the prior art, the recommendation model can use collaborative filtering technology, and the recommendation model can determine the recommended object from multiple objects to be recommended based on the association between user features and multiple objects to be recommended. However, the recommendation model recommends objects to users based on the user features of the current user, resulting in duplicate recommended object types, and is unable to recommend new objects that meet user needs, resulting in low accuracy of recommended objects. Summary of the invention

[0004] The present application provides an object recommendation method, device, equipment, storage medium and program product to solve the technical problem of low accuracy of recommended objects.

[0005] In a first aspect, the present application provides a method for recommending an object, comprising:

[0006] Acquire object operation information of a current user and multiple object information of multiple objects, wherein the object operation information is operation information of the current user on multiple historical objects within a preset time period, and the multiple objects include the multiple historical objects;

[0007] Determining an embedded feature vector of the current user according to the object operation information and the plurality of object information;

[0008] Concatenate the embedded feature vector with the user feature vector of the current user to obtain a target feature vector of the current user;

[0009] The target feature vector is processed by a preset recommendation model to obtain a recommendation object corresponding to the current user.

[0010] In a second aspect, the present application provides an object recommendation device, including an acquisition module, a determination module, a splicing processing module and a data processing module:

[0011] The acquisition module is used to acquire object operation information of a current user and multiple object information of multiple objects, wherein the object operation information is operation information of the current user on multiple historical objects within a preset period of time, and the multiple objects include the multiple historical objects;

[0012] The determination module is used to determine the embedded feature vector of the current user according to the object operation information and the plurality of object information;

[0013] The concatenation module is used to concatenate the embedded feature vector with the user feature vector of the current user to obtain the target feature vector of the current user;

[0014] The data processing module is used to perform data processing on the target feature vector through a preset recommendation model to obtain a recommendation object corresponding to the current user.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0016] The memory stores computer-executable instructions;

[0017] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0018] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0019] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0020] The object recommendation method, device, equipment, storage medium and program product provided by the present application can obtain the object operation information of the target user and the object information of multiple objects, determine the embedded feature vector of the target user based on the object operation information and the multiple object information, and determine the target feature vector of the current user based on the embedded feature vector and the user feature vector. The user's external operation information and multiple object information are added to the target feature vector, which can increase the richness of the target feature vector. By presetting the recommendation model, the target feature vector is processed and the recommended object is determined, which can improve the accuracy of the recommended object. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0022] Figure 1 A schematic diagram of an application scenario provided in an embodiment of the present application;

[0023] Figure 2 A flowchart of a method for recommending an object provided in an embodiment of the present application;

[0024] Figure 3 A flowchart of another object recommendation method provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of a coding process provided in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of a structure for determining an embedded feature vector provided in an embodiment of the present application;

[0027] Figure 6 A schematic diagram of a structure for determining a recommended feature vector provided in an embodiment of the present application;

[0028] Figure 7 A schematic diagram of the structure of a device for recommending an object provided in an embodiment of the present application;

[0029] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0030] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0031] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0032] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0033] In addition, this application involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and the use of artificial intelligence technology for automated decision-making, and a technical solution for making decisions that have a significant impact on personal rights and interests based on the results of automated decision-making. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decision-making; if the user chooses to reject, the expert decision-making process will be entered.

[0034] It should be noted that the recommended methods, devices, equipment, storage media and program products of the objects provided in this application can be used in the field of artificial intelligence technology, and can also be used in any field other than artificial intelligence technology. The application fields of the recommended methods, devices, equipment, storage media and program products of the objects in this application are not limited.

[0035] Figure 1 A schematic diagram of an application scenario provided by an embodiment of the present application. Figure 1 , may include a terminal device 101 and a server 102.

[0036] The terminal device 101 may be a mobile phone, a desktop computer, a laptop computer, a smart terminal, etc. An application program may be installed on the terminal device 101, and a service may be provided for the application program through the server 102. A user may browse an object interface in the application program through the terminal device, and the object interface may include multiple objects, and the user may click on an object to browse the specific information of the object.

[0037] The server 102 may obtain object operation information of the current user and multiple object information of multiple objects. The object operation information may be operation information of the current user on multiple historical objects within a preset period of time. The multiple objects may include multiple historical objects. The server 102 may determine the embedded feature vector of the current user based on the object operation information and the multiple object information, and may concatenate the embedded feature vector with the user feature vector of the current user to obtain the target feature vector of the current user.

[0038] The server 102 can process the target feature vector by using a preset recommendation model to obtain a recommended object corresponding to the current user. The object interface on the terminal device 101 can display the recommended objects that match the target feature vector to the user.

[0039] In the prior art, the recommendation model can use collaborative filtering technology, and the recommendation model can determine the recommended object from multiple objects to be recommended based on the association between user characteristics and multiple objects to be recommended. However, the recommendation model recommends objects to users according to the user characteristics of the current user, resulting in duplicate recommended object types, and failing to recommend new objects that meet the user characteristics, resulting in low accuracy of recommended objects.

[0040] The object recommendation method provided by the present application can obtain the object operation information of the target user and the object information of multiple objects, determine the embedded feature vector of the target user based on the object operation information and the multiple object information, and determine the target feature vector of the current user based on the embedded feature vector and the user feature vector. The user's external operation information and multiple object information are added to the target feature vector, which can increase the richness of the target feature vector. Then, by presetting the recommendation model, the target feature vector is processed and the recommended object is determined, which can improve the accuracy of the recommended object.

[0041] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0042] Figure 2 This is a flowchart of a method for recommending an object provided in an embodiment of the present application. Figure 2 , the method may include:

[0043] S201: Obtain object operation information of the current user and multiple object information of multiple objects.

[0044] The execution subject of the embodiment of the present application may be a server, or an object recommendation device set in the server. The object recommendation device may be implemented by software, or by a combination of software and hardware.

[0045] The object operation information may be the operation information of the current user on multiple historical objects within a preset period of time, wherein the operation information may include clicking to enter the detail page of the historical object, purchasing the historical object, staying at the display position of the historical object on the object interface, etc.

[0046] The preset time period may be a period from the current time to the preset time period. For example, assuming that the preset time period is 24 hours and the current time is 9:00 on December 11, 2024, the preset time period may be 9:00 on December 10, 2024 to 9:00 on December 11, 2024.

[0047] The multiple objects may be all objects in the application program, and the multiple objects may include multiple history objects.

[0048] The object information may be all descriptive information of the object. For example, the object information may include object identification, object category, object brand, object price, keywords, object attributes, object specifications, and the like.

[0049] S202: Determine an embedded feature vector of the current user according to the object operation information and the multiple object information.

[0050] Further, the object operation information is subjected to a first encoding process to obtain an operation feature vector; each object information is subjected to a second encoding process to obtain an object feature vector; and an embedded feature vector is determined according to the operation feature vector and the object feature vector.

[0051] The operation characteristic vector may be used to indicate the characteristics of multiple historical objects currently selected by the user, and the object characteristic vector may be used to indicate the characteristics of multiple objects.

[0052] By encoding the object operation information and the object information, determining the object operation information as an operation feature vector, and determining each object information as an object feature vector, it is convenient to perform feature processing on the operation feature vector and the object feature vector, and the efficiency of data processing can be improved.

[0053] S203: Concatenate the embedded feature vector with the user feature vector of the current user to obtain a target feature vector of the current user.

[0054] The user feature vector of the current user may be determined in the following manner: user information of the current user may be obtained, where the user information may include gender, age, education level, network access duration, height, weight, etc. The user information may be encoded to obtain a user feature vector.

[0055] For example, the embedded feature vector is {vector 1}, and the user feature vector is {vector 2}, then the target feature vector can be determined to be {vector 1, vector 2}.

[0056] S204: Perform data processing on the target feature vector through a preset recommendation model to obtain a recommendation object corresponding to the current user.

[0057] The preset recommendation model may be a trained neural network model, the target feature vector may be input into the preset recommendation model, the output of the preset recommendation model may be determined as a recommended feature vector, and the recommended object may be determined based on the recommended feature vector.

[0058] The recommended characteristic vector may include an object code of at least one recommended object.

[0059] Furthermore, in the preset recommendation model, the target feature vector may be sampled multiple times to obtain multiple sampled feature vectors; a multi-order feature vector may be determined based on the multiple sampled feature vectors; and a recommended object may be determined based on the multiple sampled feature vectors and the multi-order feature vectors.

[0060] Through multiple sampling processes, multiple adopted feature vectors can be determined, and then the multiple sampled feature vectors can be fused to obtain multi-order feature vectors, which can increase the diversity of feature learning and improve the effectiveness of recommended objects.

[0061] After recommending an object to the current user, if the recommended object is not the object required by the user, the user's selected object can be fed back, and the model can be updated according to the user's selected object.

[0062] The object recommendation method provided in this embodiment can obtain the object operation information of the target user and the object information of multiple objects, determine the embedded feature vector of the target user based on the object operation information and the multiple object information, and determine the target feature vector of the current user based on the embedded feature vector and the user feature vector. The user's external operation information and multiple object information are added to the target feature vector, which can increase the richness of the target feature vector. Then, by presetting the recommendation model, the target feature vector is processed and the recommended object is determined, which can improve the accuracy of the recommended object.

[0063] Figure 3 A flowchart of another method for recommending an object provided in an embodiment of the present application. Figure 3 , the method may include:

[0064] S301: Obtain object operation information of the current user and multiple object information of multiple objects.

[0065] The execution process of S301 can refer to the execution process of S201, which will not be repeated here.

[0066] S302: Perform a first encoding process on the object operation information to obtain an operation feature vector.

[0067] Optionally, the object operation information may be encoded using a first language model to obtain an operation embedding vector; and the operation embedding vector may be aggregated to obtain an operation feature vector.

[0068] The first language model may be a pre-trained large language model (LLM), and the first language model may include a decoder layer, and an operation embedding vector corresponding to the object operation information may be generated through the decoder layer of the first language model.

[0069] The operating embedding vector is the input result of the intermediate layer of the first language model.

[0070] The operation embedding vector may include eigenvalues ​​corresponding to multiple operation features, as shown in the following formula:

[0071]

[0072] in, It can be used to indicate the i-th operation feature, Encoder(·) is used to indicate encoding, and Aggr(·) is used to indicate concatenation of multiple encodings. The eigenvalue used to indicate the characteristic of the i-th operation.

[0073] In order to ensure that the operation feature vector has the same vector dimension each time the current user is processed, the operation embedding vector may be aggregated through an average pooling layer to obtain an operation feature vector, and the vector dimension of the operation feature vector is a preset dimension.

[0074] The average pooling layer can reduce the dimension of the operation embedding vector, reduce the complexity of the feature extraction process, and retain the object operation information.

[0075] The object operation information can be encoded through the first language model to obtain an operation embedding vector, and then the operation embedding vector is aggregated to obtain an operation feature vector. This can unify the dimensions of the operation feature vector, reduce the difficulty of processing, and improve the processing efficiency of object recommendation.

[0076] Figure 4 A schematic diagram of a coding process provided in an embodiment of the present application. Figure 4 The object operation information may include multiple phrase sequences, which can be processed by a large language model to obtain an operation embedding vector, and the operation embedding vector can be aggregated by an average pooling layer to obtain an embedding feature vector.

[0077] S303: Perform a second encoding process on each object information to obtain an object feature vector.

[0078] Furthermore, each object information may be encoded through a second language model to obtain an object embedding sequence; the object embedding sequence may be aggregated to obtain an object feature vector, wherein the vector dimension of the object feature vector is a second preset dimension.

[0079] The object embedding sequence can be aggregated through an average pooling layer to obtain an object feature vector, wherein the vector dimensions of the object feature vector and the operation feature vector can be different or the same.

[0080] Before performing the second encoding process on the object information, feature extraction may be performed on the object information to obtain a plurality of object features, where the object features may be object identification, object category, object brand, object price, keywords and object attributes.

[0081] A second encoding process is performed on the plurality of object characteristics to obtain an object feature vector.

[0082] The second language model may be a trained LLM model. The second language model may include a decoder layer, and the object embedding vectors corresponding to the plurality of object features may be generated by the decoder layer of the second language model.

[0083] The object embedding vector may include the eigenvalues ​​of multiple object features. For any eigenvalue, the following formula can be used:

[0084]

[0085] in, It is used to indicate the j-th object feature, Encoder(·) is used to indicate encoding, and Aggr(·) is used to indicate concatenation of multiple encodings. The feature value used to indicate the j-th object feature of the object embedding vector.

[0086] The object embedding vector can be aggregated by means of a pooling layer to obtain the object feature vector.

[0087] S304: Determine the preference feature vector of the current user according to the operation feature vector.

[0088] Furthermore, the operation feature vector can be subjected to a first analysis process through a user preference model to obtain a first preference vector; the operation feature vector can be subjected to a second analysis process through a first expert model to obtain a second preference vector; and the first preference vector and the second preference vector can be fused to obtain a preference feature vector.

[0089] The execution formula for determining the first preference vector may be as follows:

[0090]

[0091] in, The i-th eigenvalue used to indicate the first preference vector, r i p The eigenvalue used to indicate the characteristic of the i-th operation, g p (·) is used to indicate the network function of the user preference model, and Softmax(·) is used to indicate the normalization process.

[0092] The first expert model may include feature extraction of preference information and feature extraction of shared information, and the execution formula for determining the second preference vector may be as follows:

[0093]

[0094] in, The eigenvalue used to indicate the i-th preference feature of the second preference vector, r i pis used to indicate the eigenvalue of the i-th operating feature, e1(·) is used to indicate the network of the first expert model, Used to indicate the weight of the sth preference expert on the i-th preference feature, Used to indicate the number of preferred experts, Used to indicate the weight of q shared experts on the i-th preference feature, Used to indicate the number of shared experts.

[0095] The network of the first expert model may be a multi-layer perceptron.

[0096] Through the user preference model, the operation feature vector is subjected to a first analysis and processing to obtain a first preference vector, which can accurately capture the unique characteristics of the current user during operation. Through the first expert model, the operation feature vector is subjected to a first analysis and processing to obtain a second preference vector, which can integrate the environmental impact information of preference experts and shared experts. Based on the first preference vector and the second preference vector, the preference feature vector of the current user is determined, which can improve the accuracy of the recommended objects.

[0097] S305: Determine an object enhancement vector for the current user according to the object feature vector and the shared feature vector.

[0098] The object enhancement vector may be used to indicate the relevance characteristics of the current user and multiple objects.

[0099] Furthermore, the object feature vector is subjected to a first analysis process through the object expert model to obtain a first enhanced vector; the object feature vector is subjected to a second analysis process through the second expert model to obtain a second enhanced vector; the first enhanced vector and the second enhanced vector are fused to obtain an object enhanced vector.

[0100] The execution formula for determining the first enhancement vector may be as follows:

[0101]

[0102] in, is used to indicate the jth eigenvalue of the first enhanced vector, The feature value used to indicate the j-th object feature of the object embedding vector, g p (·) is used to indicate the network function of the user preference model, and Softmax(·) is used to indicate the normalization process.

[0103] The second expert model may include feature extraction of shared information and feature extraction of object experts, and the execution formula for determining the second enhancement vector may be as follows:

[0104]

[0105] in, A feature value indicating the j-th object feature of the second enhanced vector, is used to indicate the feature value of the jth object feature of the object embedding vector, e2(·) is used to indicate the network of the second expert model, Used to indicate the weight of s shared experts on the features of the i-th object, Used to indicate the number of shared experts, It is used to indicate the weight of the t-th object expert on the i-th object feature. Used to indicate the number of experts on a subject.

[0106] The network of the second expert model may be a multi-layer perceptron.

[0107] By accurately capturing the unique features of multiple objects through the object expert model and then determining the object enhancement vector, the accuracy of the recommended objects can be improved.

[0108] S306: Concatenate the preference feature vector and the object enhancement vector to obtain an embedded feature vector.

[0109] By embedding feature vectors, the relevant features between the current user's preferences and multiple objects can be determined, which can improve the accuracy of recommended objects.

[0110] Figure 5 This is a schematic diagram of a structure for determining an embedded feature vector provided in an embodiment of the present application. Figure 5 , through the user preference model, the operation feature vector is subjected to a first analysis process to obtain a first preference vector. The expert model includes a first expert model and a second expert model. Through the first expert model, the operation feature vector is subjected to a second analysis process to obtain a second preference vector; the first preference vector and the second preference vector are fused to obtain a preference feature vector.

[0111] The object feature vector can be subjected to a first analysis process by an object expert model to obtain a first enhanced vector; the object feature vector can be subjected to a second analysis process by a second expert model to obtain a second enhanced vector; the first enhanced vector and the second enhanced vector can be fused to obtain an object enhanced vector. The preference feature vector and the object enhanced vector can be concatenated to obtain an embedded feature vector.

[0112] S307: Concatenate the embedded feature vector with the user feature vector of the current user to obtain a target feature vector of the current user.

[0113] The execution process of S307 can refer to the execution process of S203, which will not be repeated here.

[0114] S308 , performing multiple sampling processes on the target feature vector to obtain multiple sampling feature vectors.

[0115] In some possible embodiments, the target feature vector may be sampled twice, namely, a first sampling and a second sampling.

[0116] The target feature vector may be sampled for the first time to obtain a first sampling feature vector. The first sampling may be a second-order feature intersection. The first sampling may refer to the following formula:

[0117]

[0118] in, is the i-th sampling value of the first sampling feature vector, x i is the i-th eigenvalue of the target eigenvector, x j is the i+1th eigenvalue of the target eigenvector, w ij is the learnable parameter after the intersection of the i-th and j-th features.

[0119] The target feature vector may be sampled a second time to obtain a second sampled feature vector. The second sampling may be a third-order feature intersection. The second sampling may refer to the following formula:

[0120]

[0121] in, is the i-th sampling value of the second sampling feature vector, x i is the i-th eigenvalue of the target eigenvector, x j is the i+1th eigenvalue of the target eigenvector, x k is the i+2th eigenvalue of the target eigenvector, w ijk is the learnable parameter after the intersection of the i-th, j-th, and k-th features of the target feature vector.

[0122] S309: Determine a multi-order feature vector according to the multiple sampled feature vectors.

[0123] Multiple sampling feature vectors may be fused to obtain a fused feature vector, and the fused feature vector has the same vector dimension as each sampling feature vector.

[0124] The fusion process may be to add the sampling values ​​at the corresponding positions of each sampling feature vector.

[0125] The fused feature vector can be subjected to global average pooling to obtain a statistical feature vector, and then the statistical feature vector can be subjected to dimensionality reduction to obtain a multi-order feature vector.

[0126] Perform global average pooling on the fusion feature vector to obtain the statistical feature vector Z = [z1…, z i ··, z f ], where i∈[1,…,f], z i The scalar value representing the global information of the i-th feature intersection, f is the dimension of the statistical feature vector, which can be seen in the following formula:

[0127]

[0128] Among them, F gp (·) is used to indicate global average pooling processing, It is used to indicate the tth value of the eigenvalue corresponding to the i-th feature of the fused feature vector, and K is used to indicate the number of bits of each eigenvalue.

[0129] Perform dimension reduction on the statistical characteristic vector to obtain a multi-order characteristic vector S, which can be seen in the following formula:

[0130] S=F fc (Z) = σ1(w1Z)

[0131] Among them, F fc (·) is used to indicate the dimensionality reduction process, σ1(·) is the activation function, and w1 is a learnable parameter.

[0132] S310: Determine a recommendation object according to a plurality of sampled feature vectors and a multi-order feature vector.

[0133] The first weight corresponding to each adopted feature vector and the second weight corresponding to the multi-order feature vector can be determined; data processing is performed based on the first weight corresponding to each adopted feature vector, each adopted feature vector, the multi-order feature vector, and the second weight to obtain a recommended feature vector for the current user; and the recommended object is determined based on the recommended feature vector.

[0134] The first weight corresponding to each adopted eigenvector and the second weight corresponding to the multi-order eigenvectors may be determined by a normalization (softmax) operator.

[0135] For example, assuming there are two sub-sampling feature vectors, namely the first sampling feature vector U1 and the second sampling feature vector U2, the multi-order feature vector U3, and the first weight a1 of the first sampling feature vector U1, can be seen in the following formula:

[0136]

[0137] The recommended feature vector V can be determined by the following formula:

[0138] V=a1×U1+a2×U2+a3×U3

[0139] Among them, a1 is the first weight of the first sampling feature vector U1, a2 is the first weight of the second sampling feature vector U2, and a3 is the second weight of the multi-order feature vector U3.

[0140] The recommended characteristic vector may include an object identification of at least one object, and the recommended object may be determined according to the object identification of the at least one object.

[0141] Through feature crossover, key features can be dynamically added, the weights between feature vectors can be enhanced to enhance interactions, and the proportion of feature combinations with smaller contributions can be compressed, thereby improving the accuracy of recommended objects.

[0142] Figure 6 This is a schematic diagram of the structure of determining the recommended feature vector provided in the embodiment of the present application. Figure 6 , the target feature vector is sampled twice respectively, and a first sampling feature vector and a second sampling feature vector can be obtained. The first sampling feature vector and the second sampling feature vector can be fused to obtain a fused feature vector. The fused feature vector can be globally averaged pooled to obtain a statistical feature vector, and then the statistical feature vector can be dimensionally reduced to obtain a multi-order feature vector. The recommended feature vector of the current user can be obtained according to the first adopted feature vector, the second adopted feature vector, and the multi-order feature vector.

[0143] The object recommendation method provided in the embodiment of the present application adds the user's external operation information and multiple object information to the target feature vector, which can increase the richness of the target feature vector. Then, by presetting the recommendation model, the target feature vector is processed and the recommended object is determined, which can improve the accuracy of the recommended object.

[0144] The above embodiment can be implemented through an adaptive recommendation model. The adaptive recommendation model can process the object operation information and multiple object information to obtain a target feature vector. The adaptive recommendation model can also process the target feature vector through a preset recommendation model to determine the recommended object.

[0145] The adaptive recommendation model can be obtained by training the constructed initial recommendation model through the training set. The learnable parameters in the above embodiment can be obtained according to the training. The training set may include multiple training samples, and the multiple training samples may include positive samples and negative samples.

[0146] Positive samples can be composed of the embedded feature vectors and user feature vectors of the same user at different times. Positive samples can reflect the actual preferences of users. Negative samples can be composed of the embedded feature vectors and user feature vectors of different users. Negative samples can reflect the non-preferences of users. By increasing the diversity of the sample set by positive and negative samples, the model can learn to distinguish samples.

[0147] Figure 7 This is a schematic diagram of a structure of an object recommendation device provided in an embodiment of the present application. Figure 7 The object recommendation device 10 may include an acquisition module 11, a determination module 12, a splicing processing module 13 and a data processing module 14:

[0148] The acquisition module 11 is used to acquire object operation information of the current user and multiple object information of multiple objects, where the object operation information is operation information of the current user on multiple historical objects within a preset period of time, and the multiple objects include multiple historical objects;

[0149] The determination module 12 is used to determine the embedded feature vector of the current user according to the object operation information and the multiple object information;

[0150] The concatenation module 13 is used to concatenate the embedded feature vector with the user feature vector of the current user to obtain the target feature vector of the current user;

[0151] The data processing module 14 is used to perform data processing on the target feature vector through a preset recommendation model to obtain a recommendation object corresponding to the current user.

[0152] The object recommendation device provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effect are similar, which will not be described in detail in this embodiment.

[0153] In some possible embodiments, the determination module 12 is specifically used for:

[0154] Performing a first encoding process on the object operation information to obtain an operation characteristic vector, where the operation characteristic vector is used to indicate characteristics of a plurality of historical objects;

[0155] Performing a second encoding process on each object information to obtain an object feature vector, where the object feature vector is used to indicate features of multiple objects;

[0156] An embedding feature vector is determined according to the operation feature vector and the object feature vector.

[0157] In some possible embodiments, the determination module 12 is specifically used for:

[0158] Through the first language model, the object operation information is encoded to obtain the operation embedding vector;

[0159] Aggregation processing is performed on the operation embedding vector to obtain an operation feature vector, where the vector dimension of the operation feature vector is a first preset dimension.

[0160] In some possible embodiments, the determination module 12 is specifically used for:

[0161] Determine the current user's preference feature vector based on the operation feature vector;

[0162] Determine an object enhancement vector of the current user according to the object feature vector, where the object enhancement vector is used to indicate the correlation characteristics between the current user and the multiple objects;

[0163] The preference feature vector and the object enhancement vector are concatenated to obtain an embedded feature vector.

[0164] In some possible embodiments, the determination module 12 is specifically used for:

[0165] Performing a first analysis and processing on the operation feature vector through the user preference model to obtain a first preference vector;

[0166] Performing a second analysis process on the operation feature vector through the first expert model to obtain a second preference vector;

[0167] The first preference vector and the second preference vector are fused to obtain a preference feature vector.

[0168] In some possible embodiments, the determination module 12 is specifically used for:

[0169] Performing a first analysis process on the object feature vector through the object expert model to obtain a first enhanced vector;

[0170] Performing a second analysis process on the object feature vector through a second expert model to obtain a second enhanced vector;

[0171] The first enhancement vector and the second enhancement vector are fused to obtain an object enhancement vector.

[0172] In some possible embodiments, the data processing module 14 is specifically used for:

[0173] Performing multiple sampling processes on the target feature vector to obtain multiple sampling feature vectors;

[0174] Determine a multi-order feature vector according to a plurality of sampled feature vectors;

[0175] The recommended object is determined according to multiple sampling feature vectors and multi-order feature vectors.

[0176] In some possible embodiments, the data processing module 14 is specifically used for:

[0177] Determine a first weight corresponding to each adopted eigenvector and a second weight corresponding to a multi-order eigenvector;

[0178] Performing data processing according to the first weight corresponding to each adopted feature vector, each adopted feature vector, the multi-order feature vector, and the second weight to obtain a recommendation feature vector for the current user;

[0179] According to the recommended feature vector, the recommended object is determined.

[0180] The object recommendation device provided in this embodiment can execute the method provided in the above method embodiment, and its implementation principle and technical effect are similar, which will not be described in detail in this embodiment.

[0181] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 8 The electronic device 20 may include a processor 21 and a memory 22. Exemplarily, the processor 21 and the memory 22 are interconnected via a bus 23.

[0182] The memory 22 stores computer executable instructions;

[0183] The processor 21 executes the computer-executable instructions stored in the memory 22, so that the processor 21 executes the object recommendation method shown in the above method embodiment.

[0184] Accordingly, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the recommended method for implementing the object of the above method embodiment is used.

[0185] Accordingly, an embodiment of the present application may also provide a computer program product, including a computer program, which, when executed by a processor, can implement the object recommendation method shown in the above method embodiment.

[0186] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0188] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0190] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0191] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0192] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0193] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0194] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for recommending an object, characterized in that: include: Acquire object operation information of a current user and multiple object information of multiple objects, wherein the object operation information is operation information of the current user on multiple historical objects within a preset time period, and the multiple objects include the multiple historical objects; Determining an embedded feature vector of the current user according to the object operation information and the plurality of object information; Concatenate the embedded feature vector with the user feature vector of the current user to obtain a target feature vector of the current user; The target feature vector is processed by a preset recommendation model to obtain a recommendation object corresponding to the current user.

2. The method according to claim 1, characterized in that Determining the embedded feature vector of the current user according to the object operation information and the plurality of object information includes: Performing a first encoding process on the object operation information to obtain an operation feature vector, wherein the operation feature vector is used to indicate features of the plurality of historical objects; Performing a second encoding process on each object information to obtain an object feature vector, wherein the object feature vector is used to indicate features of the multiple objects; The embedded feature vector is determined according to the operation feature vector and the object feature vector.

3. The method according to claim 2, characterized in that Performing a first encoding process on the object operation information to obtain an operation feature vector includes: encoding the object operation information through a first language model to obtain an operation embedding vector; Aggregation processing is performed on the operation embedding vector to obtain the operation feature vector, where the vector dimension of the operation feature vector is a first preset dimension.

4. The method according to claim 2, characterized in that: Determining the embedded feature vector according to the operation feature vector and the object feature vector includes: Determining a preference feature vector of the current user according to the operation feature vector; Determining an object enhancement vector of the current user according to the object feature vector, wherein the object enhancement vector is used to indicate a correlation feature between the current user and the multiple objects; The preference feature vector and the object enhancement vector are concatenated to obtain the embedded feature vector.

5. The method according to claim 4, characterized in that Determining the preference feature vector of the current user according to the operation feature vector includes: Performing a first analysis process on the operation feature vector using a user preference model to obtain a first preference vector; Performing a second analysis process on the operation feature vector through the first expert model to obtain a second preference vector; The first preference vector and the second preference vector are fused to obtain the preference feature vector.

6. The method according to claim 4, characterized in that Determining the object enhancement vector of the current user according to the object feature vector includes: Performing a first analysis process on the object feature vector through an object expert model to obtain a first enhanced vector; Performing a second analysis process on the object feature vector by a second expert model to obtain a second enhanced vector; The first enhancement vector and the second enhancement vector are fused to obtain the object enhancement vector.

7. The method according to any one of claims 1 to 6, characterized in that: Performing data processing on the target feature vector to obtain a recommended object corresponding to the current user includes: Performing multiple sampling processes on the target feature vector to obtain multiple sampling feature vectors; Determine a multi-order feature vector according to the multiple sampled feature vectors; The recommended object is determined according to the multiple sampled feature vectors and the multi-order feature vectors.

8. The method according to claim 7, characterized in that Determining the recommended object according to the multiple adopted feature vectors and the multi-order feature vectors includes: Determine a first weight corresponding to each adopted eigenvector and a second weight corresponding to a multi-order eigenvector; Performing data processing according to the first weight corresponding to each adopted feature vector, each adopted feature vector, the multi-order feature vector, and the second weight to obtain a recommendation feature vector for the current user; The recommended object is determined according to the recommended characteristic vector.

9. An object recommendation device, characterized in that: Including acquisition module, determination module, splicing processing module and data processing module: The acquisition module is used to acquire object operation information of a current user and multiple object information of multiple objects, wherein the object operation information is operation information of the current user on multiple historical objects within a preset period of time, and the multiple objects include the multiple historical objects; The determination module is used to determine the embedded feature vector of the current user according to the object operation information and the plurality of object information; The concatenation module is used to concatenate the embedded feature vector with the user feature vector of the current user to obtain the target feature vector of the current user; The data processing module is used to perform data processing on the target feature vector through a preset recommendation model to obtain a recommendation object corresponding to the current user.

10. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

12. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 8 when being executed by a processor.

Images