Resource recommendation method, resource recommendation model training method, device and equipment

By filtering related data from the interaction data between the target domain and the source domain, and combining object and resource characteristics, the accuracy of resource recommendations is improved, and the problem of inaccurate recommendations caused by the sparsity of target domain data is solved.

CN119003873APending Publication Date: 2024-11-22TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411097636.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In existing technologies, the accuracy of resource recommendations is low due to the limited feedback data from users in a particular field.

Method used

By acquiring interaction data of objects in the target domain and source domain, filtering out source domain interaction data associated with the target domain, and combining object, resource, target domain, and source domain interaction data, the recommendation prediction results for resources are determined.

Benefits of technology

It improves the accuracy of resource recommendations, compensates for the lack of interactive data in the target domain, and reduces the interference of irrelevant data on prediction results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119003873A_ABST
    Figure CN119003873A_ABST
Patent Text Reader

Abstract

The invention discloses a resource recommendation method, a resource recommendation model training method, a resource recommendation model training device and equipment, and belongs to the technical field of computers. The method comprises the steps of obtaining object data of an object and resource data of a target resource, wherein the target resource belongs to a target domain; first source domain interaction data of the object and target domain interaction data of the object are obtained, the first source domain interaction data comprise a plurality of source domain interaction items, and the target domain interaction data comprise a plurality of target domain interaction item source domains and target domains which belong to different fields; in the first source domain interaction data, at least one source domain interaction item associated with the target domain is determined, and the at least one source domain interaction item associated with the target domain forms second source domain interaction data; and based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, determining a recommendation prediction result of the target resource. Due to the introduction of the source domain interaction data, the defect of insufficient target domain interaction data can be made up, and the accuracy of resource recommendation can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a resource recommendation method, a training method, a device and a equipment for a resource recommendation model. Background Art

[0002] With the rapid development of computer technology, artificial intelligence technology is being used more and more widely in the field of recommendation. For example, it can recommend items, videos, music and other resources to users. Users will give feedback on the resources they receive, and the feedback data can be used as the basis for the recommendation system to make recommendations to users.

[0003] In the related art, for a recommendation task in a certain field, the resources in the field are usually inferred to be recommended to the user based on the feedback data generated by the user in the field. However, if the feedback data generated by the user in the field is small, the amount of data based on which the resource recommendation is made is small, which leads to low accuracy of the resource recommendation. Summary of the invention

[0004] The embodiment of the present application provides a resource recommendation method, a resource recommendation model training method, a device and a device, which can improve the accuracy of resource recommendation. The technical solution is as follows:

[0005] In one aspect, a resource recommendation method is provided, the method comprising:

[0006] Obtaining object data of an object and resource data of a target resource, the target resource belonging to a target domain;

[0007] Acquire first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction items, wherein the source domain interaction items refer to the interaction between the object and resources in the source domain, and the target domain interaction data includes a plurality of target domain interaction items, wherein the target domain interaction items refer to the interaction between the object and resources in the target domain, and the source domain and the target domain belong to different fields;

[0008] In the first source domain interaction data, at least one source domain interaction item associated with the target domain is determined, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data;

[0009] Based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, a recommendation prediction result of the target resource is determined, and the recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0010] Optionally, if the similarity between the entity to which the resource in the source domain interactive event belongs and any of the target entities is greater than a similarity threshold, determining that the source domain interactive event is associated with the target domain includes:

[0011] Obtaining target entity features of each of the target entities;

[0012] Extracting features of entities to which resources in the source domain interactive events belong, and obtaining source domain entity features;

[0013] If the similarity between the source domain entity feature and any of the target entity features is greater than the similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0014] Optionally, the feature extraction sub-model further includes a second attention network, and the second attention network includes a second mapping matrix; and determining the source domain interaction feature and the target domain interaction feature based on the second fused interaction feature includes:

[0015] Separating the intermediate source domain interaction features from the second fused interaction features;

[0016] By means of the second attention network, the resource feature is multiplied by the second mapping matrix to obtain a second query feature, and the intermediate source domain interaction feature is multiplied by the second mapping matrix to obtain a second key feature and a second value feature;

[0017] The second query feature, the second key feature and the second value feature are fused through the second attention network to obtain the source domain interaction feature.

[0018] Optionally, the feature extraction sub-model further includes a third attention network, and the third attention network includes a third mapping matrix; and determining the source domain interaction feature and the target domain interaction feature based on the second fused interaction feature includes:

[0019] Separating an intermediate target domain interaction feature from the second fused interaction feature;

[0020] Through the third attention network, multiplying the resource feature with the third mapping matrix to obtain a third query feature, and multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain a third key feature and a third value feature;

[0021] The third query feature, the third key feature and the third value feature are fused through the third attention network to obtain the target domain interaction feature.

[0022] Optionally, after determining the recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, the method further includes:

[0023] If the recommendation prediction result meets the recommendation condition, the target resource is recommended to the object.

[0024] On the other hand, a method for training a resource recommendation model is provided, the method comprising:

[0025] Acquire sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object;

[0026] Acquire first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to interactions between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to interactions between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains;

[0027] In the first sample source domain interaction data, at least one sample source domain interaction event associated with the target domain is determined, and the at least one sample source domain interaction event associated with the target domain constitutes the second sample source domain interaction data;

[0028] Determine, by means of a resource recommendation model, a recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data, and the target domain sample interaction data, wherein the recommendation prediction result is used to characterize the predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object;

[0029] The resource recommendation model is trained based on the recommendation prediction result and the sample label.

[0030] Optionally, the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, and the first attention network includes a first mapping matrix; the feature extraction sub-model is used to extract features from the second sample source domain interaction data and the sample target domain interaction data to obtain sample source domain interaction features and sample target domain interaction features, including:

[0031] Through the feature extraction network, feature extraction is performed on the second sample source domain interaction data to obtain initial sample source domain interaction features, and feature extraction is performed on the sample target domain interaction data to obtain initial sample target domain interaction features;

[0032] The initial sample source domain interaction feature and the initial sample target domain interaction feature are fused through the feature fusion network to obtain a first sample fused interaction feature;

[0033] Through the first attention network, multiplying the first sample fusion interaction feature by the first mapping matrix to obtain a first sample query feature, a first sample key feature and a first sample value feature, and fusing the first sample query feature, the first sample key feature and the first sample value feature to obtain a second sample fusion interaction feature;

[0034] Based on the second sample fusion interaction feature, the sample source domain interaction feature and the sample target domain interaction feature are determined.

[0035] Optionally, the feature extraction sub-model further includes a second attention network, and the second attention network includes a second mapping matrix; and determining the sample source domain interaction feature and the sample target domain interaction feature based on the second sample fusion interaction feature includes:

[0036] Separating the intermediate sample source domain interaction feature from the second sample fusion interaction feature;

[0037] Through the second attention network, the sample resource feature is multiplied by the second mapping matrix to obtain a second sample query feature, and the intermediate sample source domain interaction feature is multiplied by the second mapping matrix to obtain a second sample key feature and a second sample value feature;

[0038] The second sample query feature, the second sample key feature and the second sample value feature are fused through the second attention network to obtain the sample source domain interaction feature.

[0039] Optionally, the feature extraction sub-model further includes a third attention network, and the third attention network includes a third mapping matrix; the determining the sample source domain interaction feature and the sample target domain interaction feature based on the second sample fusion interaction feature includes:

[0040] Separating the intermediate sample target domain interaction feature from the second sample fusion interaction feature;

[0041] Through the third attention network, the sample resource feature is multiplied by the third mapping matrix to obtain a third sample query feature, and the intermediate sample target domain interaction feature is multiplied by the third mapping matrix to obtain a third sample key feature and a third sample value feature;

[0042] The third sample query feature, the third sample key feature and the third sample value feature are fused through the third attention network to obtain the sample target domain interaction feature.

[0043] Optionally, determining at least one sample source domain interaction item associated with the target domain in the first sample source domain interaction data includes:

[0044] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0045] For any sample source domain interaction event in the first sample source domain interaction data, if the entity to which the resource in the sample source domain interaction event belongs is any of the target entities, it is determined that the sample source domain interaction event is associated with the target domain.

[0046] Optionally, determining at least one sample source domain interaction item associated with the target domain in the first sample source domain interaction data includes:

[0047] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0048] For any sample source domain interaction event in the first sample source domain interaction data, if the similarity between the entity to which the resource in the sample source domain interaction event belongs and any of the target entities is greater than a similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0049] Optionally, if the similarity between the entity to which the resource in the sample source domain interactive event belongs and any of the target entities is greater than a similarity threshold, determining that the sample source domain interactive event is associated with the target domain includes:

[0050] Obtaining target entity features of each of the target entities;

[0051] Extracting features of entities to which resources belong in the sample source domain interactive events to obtain sample source domain entity features;

[0052] If the similarity between the sample source domain entity feature and any of the target entity features is greater than the similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0053] In another aspect, a resource recommendation device is provided, the device comprising:

[0054] A first acquisition module, configured to acquire object data of an object and resource data of a target resource, wherein the target resource belongs to a target domain;

[0055] A second acquisition module is provided for acquiring first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction items, wherein the source domain interaction items refer to the interaction between the object and resources in the source domain, and the target domain interaction data includes a plurality of target domain interaction items, wherein the target domain interaction items refer to the interaction between the object and resources in the target domain, and the source domain and the target domain belong to different domains;

[0056] a screening module, configured to determine, in the first source domain interaction data, at least one source domain interaction event associated with the target domain, wherein the at least one source domain interaction event associated with the target domain constitutes second source domain interaction data;

[0057] A prediction module is used to determine a recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, wherein the recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0058] Optionally, the screening module is used to:

[0059] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0060] For any source domain interaction event in the first source domain interaction data, if the entity to which the resource in the source domain interaction event belongs is any of the target entities, it is determined that the source domain interaction event is associated with the target domain.

[0061] Optionally, the screening module is used to:

[0062] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0063] For any source domain interaction event in the first source domain interaction data, if the similarity between the entity to which the resource in the source domain interaction event belongs and any of the target entities is greater than a similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0064] Optionally, the screening module is used to:

[0065] Obtaining target entity features of each of the target entities;

[0066] Extracting features of entities to which resources in the source domain interactive events belong, and obtaining source domain entity features;

[0067] If the similarity between the source domain entity feature and any of the target entity features is greater than the similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0068] Optionally, the prediction module is used to:

[0069] Extracting features from the object data to obtain object features;

[0070] Extracting features from the resource data to obtain resource features;

[0071] Performing feature extraction on the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features;

[0072] Based on the object features, the resource features, the source domain interaction features, and the target domain interaction features, a recommendation prediction result of the target resource is determined.

[0073] Optionally, the prediction module is used to:

[0074] Performing feature extraction on the second source domain interaction data to obtain initial source domain interaction features, and performing feature extraction on the target domain interaction data to obtain initial target domain interaction features;

[0075] Fusing the initial source domain interaction feature and the initial target domain interaction feature to obtain a first fused interaction feature;

[0076] Multiplying the first fused interaction feature by the first mapping matrix to obtain a first query feature, a first key feature, and a first value feature;

[0077] Fusing the first query feature, the first key feature, and the first value feature to obtain a second fused interaction feature;

[0078] Based on the second fused interaction feature, the source domain interaction feature and the target domain interaction feature are determined.

[0079] Optionally, the prediction module is used to:

[0080] Separating the intermediate source domain interaction features from the second fused interaction features;

[0081] Multiplying the resource feature by a second mapping matrix to obtain a second query feature;

[0082] Multiplying the intermediate source domain interaction feature by the second mapping matrix to obtain a second key feature and a second value feature;

[0083] The second query feature, the second key feature and the second value feature are fused to obtain the source domain interaction feature.

[0084] Optionally, the prediction module is used to:

[0085] Separating an intermediate target domain interaction feature from the second fused interaction feature;

[0086] Multiplying the resource feature by a third mapping matrix to obtain a third query feature;

[0087] Multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain a third key feature and a third value feature;

[0088] The third query feature, the third key feature and the third value feature are fused to obtain the target domain interaction feature.

[0089] Optionally, the resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model, and the prediction module is used to:

[0090] Performing feature extraction on the object data through the feature extraction sub-model to obtain object features;

[0091] By using the feature extraction sub-model, feature extraction is performed on the resource data to obtain resource features;

[0092] By using the feature extraction sub-model, feature extraction is performed on the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features;

[0093] The recommendation prediction sub-model is used to determine the recommendation prediction result of the target resource based on the object feature, the resource feature, the source domain interaction feature and the target domain interaction feature.

[0094] Optionally, the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, wherein the first attention network includes a first mapping matrix; and the prediction module is used to:

[0095] Through the feature extraction network, feature extraction is performed on the second source domain interaction data to obtain initial source domain interaction features, and feature extraction is performed on the target domain interaction data to obtain initial target domain interaction features;

[0096] The initial source domain interaction feature and the initial target domain interaction feature are fused through the feature fusion network to obtain a first fused interaction feature;

[0097] multiplying the first fused interactive feature by the first mapping matrix through the first attention network to obtain a first query feature, a first key feature, and a first value feature, and fusing the first query feature, the first key feature, and the first value feature to obtain a second fused interactive feature;

[0098] Based on the second fused interaction feature, the source domain interaction feature and the target domain interaction feature are determined.

[0099] Optionally, the feature extraction sub-model further includes a second attention network, wherein the second attention network includes a second mapping matrix; and the prediction module is used to:

[0100] Separating the intermediate source domain interaction features from the second fused interaction features;

[0101] By means of the second attention network, the resource feature is multiplied by the second mapping matrix to obtain a second query feature, and the intermediate source domain interaction feature is multiplied by the second mapping matrix to obtain a second key feature and a second value feature;

[0102] The second query feature, the second key feature and the second value feature are fused through the second attention network to obtain the source domain interaction feature.

[0103] Optionally, the feature extraction sub-model further includes a third attention network, and the third attention network includes a third mapping matrix; the prediction module is used to:

[0104] Separating an intermediate target domain interaction feature from the second fused interaction feature;

[0105] Through the third attention network, multiplying the resource feature with the third mapping matrix to obtain a third query feature, and multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain a third key feature and a third value feature;

[0106] The third query feature, the third key feature and the third value feature are fused through the third attention network to obtain the target domain interaction feature.

[0107] Optionally, the device further comprises:

[0108] A recommendation module is used to recommend the target resource to the object if the recommendation prediction result meets the recommendation condition.

[0109] In another aspect, a resource recommendation device is provided, the device comprising:

[0110] A first acquisition module, configured to acquire sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object;

[0111] A second acquisition module is used to acquire first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to the interaction between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to the interaction between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains;

[0112] a screening module, configured to determine, in the first sample source domain interaction data, at least one sample source domain interaction event associated with the target domain, wherein the at least one sample source domain interaction event associated with the target domain constitutes the second sample source domain interaction data;

[0113] A prediction module, configured to determine a recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data, and the target domain sample interaction data through a resource recommendation model, wherein the recommendation prediction result is used to characterize a predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object;

[0114] A training module is used to train the resource recommendation model based on the recommendation prediction result and the sample label.

[0115] Optionally, the resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model; the prediction module is used to:

[0116] Performing feature extraction on the sample object data through the feature extraction sub-model to obtain sample object features;

[0117] By using the feature extraction sub-model, feature extraction is performed on the sample resource data to obtain sample resource features;

[0118] By using the feature extraction sub-model, feature extraction is performed on the second sample source domain interaction data and the sample target domain interaction data to obtain sample source domain interaction features and sample target domain interaction features;

[0119] The recommendation prediction result is determined through the recommendation prediction sub-model based on the sample object characteristics, the sample resource characteristics, the sample source domain interaction characteristics and the sample target domain interaction characteristics.

[0120] Optionally, the sample source domain interaction event includes a source domain resource and an interaction time point of the source domain resource, the sample target domain interaction event includes a target domain resource and an interaction time point of the target domain resource, the sample source domain interaction feature includes a source domain resource feature of the source domain resource in each sample source domain interaction event, and the sample target domain interaction feature includes a target domain resource feature of the target domain resource in each sample target domain interaction event; the device further includes:

[0121] a first determination module, configured to determine, for any source domain resource in the second sample source domain interaction data, a first target domain resource in the sample target domain interaction data, determine a source domain resource feature of the source domain resource and a target domain resource feature of the first target domain resource as a first positive sample pair, wherein an interaction time point of the first target domain resource is closest to an interaction time point of the source domain resource;

[0122] The training module is used to:

[0123] The resource recommendation model is trained based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair.

[0124] Optionally, the device further comprises:

[0125] A second determination module is configured to determine, for any source domain resource in the second sample source domain interaction data, a second target domain resource in the sample target domain interaction data of any object other than the sample object, and determine a source domain resource feature of the source domain resource and a target domain resource feature of the second target domain resource as a first negative sample pair, wherein the second target domain resource is any target domain resource in the sample target domain interaction data of the any object;

[0126] The training module is used to:

[0127] The resource recommendation model is trained based on the recommendation prediction result, the sample label, the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair, and the similarity between the source domain resource feature and the target domain resource feature in the first negative sample pair.

[0128] Optionally, the sample source domain interaction items include source domain resources and entities to which the source domain resources belong, the sample target domain interaction items include target domain resources and entities to which the target domain resources belong, the sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction item, and the sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction item; the device further includes:

[0129] a third determination module, configured to determine, for any source domain resource in the second sample source domain interaction data, a third target domain resource in the sample target domain interaction data, and determine a source domain resource feature of the source domain resource and a target domain resource feature of the third target domain resource as a second positive sample pair, wherein the entity to which the third target domain resource belongs is the same as the entity to which the source domain resource belongs;

[0130] The training module is used to:

[0131] The resource recommendation model is trained based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the second positive sample pair.

[0132] Optionally, the device further comprises:

[0133] a fourth determination module, configured to determine a fourth target domain resource for any source domain resource in the second sample source domain interaction data, and determine a source domain resource feature of the source domain resource and a target domain resource feature of the fourth target domain resource as a second negative sample pair, wherein the fourth target domain resource is any target domain resource different from the entity to which the source domain resource belongs;

[0134] The training module is used to:

[0135] The resource recommendation model is trained based on the recommendation prediction result, the sample label, the similarity between the source domain resource features and the target domain resource features in the second positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the second negative sample pair.

[0136] Optionally, the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, wherein the first attention network includes a first mapping matrix; and the prediction module is used to:

[0137] Through the feature extraction network, feature extraction is performed on the second sample source domain interaction data to obtain initial sample source domain interaction features, and feature extraction is performed on the sample target domain interaction data to obtain initial sample target domain interaction features;

[0138] The initial sample source domain interaction feature and the initial sample target domain interaction feature are fused through the feature fusion network to obtain a first sample fused interaction feature;

[0139] Through the first attention network, multiplying the first sample fusion interaction feature by the first mapping matrix to obtain a first sample query feature, a first sample key feature and a first sample value feature, and fusing the first sample query feature, the first sample key feature and the first sample value feature to obtain a second sample fusion interaction feature;

[0140] Based on the second sample fusion interaction feature, the sample source domain interaction feature and the sample target domain interaction feature are determined.

[0141] Optionally, the feature extraction sub-model further includes a second attention network, wherein the second attention network includes a second mapping matrix; and the prediction module is used to:

[0142] Separating the intermediate sample source domain interaction feature from the second sample fusion interaction feature;

[0143] Through the second attention network, the sample resource feature is multiplied by the second mapping matrix to obtain a second sample query feature, and the intermediate sample source domain interaction feature is multiplied by the second mapping matrix to obtain a second sample key feature and a second sample value feature;

[0144] The second sample query feature, the second sample key feature and the second sample value feature are fused through the second attention network to obtain the sample source domain interaction feature.

[0145] Optionally, the feature extraction sub-model further includes a third attention network, and the third attention network includes a third mapping matrix; the prediction module is used to:

[0146] Separating the intermediate sample target domain interaction feature from the second sample fusion interaction feature;

[0147] Through the third attention network, the sample resource feature is multiplied by the third mapping matrix to obtain a third sample query feature, and the intermediate sample target domain interaction feature is multiplied by the third mapping matrix to obtain a third sample key feature and a third sample value feature;

[0148] The third sample query feature, the third sample key feature and the third sample value feature are fused through the third attention network to obtain the sample target domain interaction feature.

[0149] Optionally, the screening module is used to:

[0150] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0151] For any sample source domain interaction event in the first sample source domain interaction data, if the entity to which the resource in the sample source domain interaction event belongs is any of the target entities, it is determined that the sample source domain interaction event is associated with the target domain.

[0152] Optionally, the screening module is used to:

[0153] Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources;

[0154] For any sample source domain interaction event in the first sample source domain interaction data, if the similarity between the entity to which the resource in the sample source domain interaction event belongs and any of the target entities is greater than a similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0155] Optionally, the screening module is used to:

[0156] Obtaining target entity features of each of the target entities;

[0157] Extracting features of entities to which resources belong in the sample source domain interactive events to obtain sample source domain entity features;

[0158] If the similarity between the sample source domain entity feature and any of the target entity features is greater than the similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0159] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource recommendation method described in the above aspects, or to implement the operations performed by the resource recommendation model training method described in the above aspects.

[0160] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the resource recommendation method as described in the above aspects, or to implement the operations performed by the training method of the resource recommendation model as described in the above aspects.

[0161] On the other hand, a computer program product is provided, including a computer program, which is loaded and executed by a processor to implement the operations performed by the resource recommendation method as described in the above aspects, or to implement the operations performed by the resource recommendation model training method as described in the above aspects.

[0162] The solution provided by the embodiment of the present application, when determining whether to recommend a target resource in the target domain for an object, needs to refer to the historical interaction of the object with other resources to infer the object's interaction tendency with the target resource. Therefore, the target domain interaction data of the object in the target domain is obtained. Considering that the target domain interaction data may have insufficient data, the source domain interaction data of the object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the source domain interaction data. Then, the target domain interaction data and the screened source domain interaction data are used as references to predict the interaction between the object and the target resource if the target resource is recommended to the object. On the one hand, since the source domain interaction data is introduced, the defect of insufficient target domain interaction data can be compensated, and the impact caused by the sparse target domain interaction data is alleviated. On the other hand, since only the source domain interaction data associated with the target domain is considered, the source domain interaction data irrelevant to the target domain is discarded, so the interference of the source domain interaction data irrelevant to the target domain on the prediction result can be reduced, thereby improving the accuracy of resource recommendation as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0163] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0164] Figure 1 It is a schematic diagram of an implementation environment provided by an embodiment of the present application;

[0165] Figure 2 is a flow chart of a resource recommendation method provided in an embodiment of the present application;

[0166] Figure 3 is a flow chart of another resource recommendation method provided by an embodiment of the present application;

[0167] Figure 4 This is a data volume comparison chart provided in an embodiment of the present application;

[0168] Figure 5 is a schematic diagram of a resource recommendation method provided in an embodiment of the present application;

[0169] Figure 6is a flow chart of another resource recommendation method provided by an embodiment of the present application;

[0170] Figure 7 is a flow chart of a method for training a resource recommendation model provided in an embodiment of the present application;

[0171] Figure 8 is a flowchart of another method for training a resource recommendation model provided in an embodiment of the present application;

[0172] Fig. 9 is a schematic diagram of a training method for a resource recommendation model provided in an embodiment of the present application;

[0173] Fig.10 This is an overall architecture diagram of a resource recommendation method provided in an embodiment of the present application;

[0174] Fig.11 is a flow chart of another resource recommendation method provided by an embodiment of the present application;

[0175] Fig.12 is a schematic diagram of a service interface provided in an embodiment of the present application;

[0176] Fig.13 is a schematic diagram of a recharge interface provided in an embodiment of the present application;

[0177] Fig.14 is a schematic diagram of another service interface provided in an embodiment of the present application;

[0178] Fig.15 is a structural diagram of a resource recommendation device provided in an embodiment of the present application;

[0179] Fig.16 is a structural diagram of another resource recommendation device provided in an embodiment of the present application;

[0180] Fig.17 It is a structural diagram of a training device for a resource recommendation model provided in an embodiment of the present application;

[0181] Fig.18 It is a structural diagram of another training device for a resource recommendation model provided in an embodiment of the present application;

[0182] Fig.19 is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0183] Fig. 20 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0184] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0185] It is understood that the terms "first", "second", etc. used in this application can be used in this article to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of this application, the first source domain interaction data can be referred to as the second source domain interaction data, and similarly, the second source domain interaction data can be referred to as the first source domain interaction data.

[0186] Among them, at least one refers to one or more than one, for example, at least one source domain interaction item can be one source domain interaction item, two source domain interaction items, three source domain interaction items, and any other source domain interaction items that are an integer greater than or equal to one. Multiple refers to two or more than two, for example, multiple source domain interaction items can be two source domain interaction items, three source domain interaction items, and any other source domain interaction items that are an integer greater than or equal to two. Each refers to each of at least one, for example, each source domain interaction item refers to each source domain interaction item in multiple source domain interaction items. If the multiple source domain interaction items are 3 source domain interaction items, then each source domain interaction item refers to each source domain interaction item in the 3 source domain interaction items.

[0187] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) involved in this application are all fully authorized by users or relevant parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0188] For example, the object data, resource data, source domain interaction data, and target domain interaction data involved in this application are all fully authorized by users or relevant parties, and the collection, use, and processing of relevant data must comply with relevant laws, regulations, and standards of relevant countries and regions.

[0189] The resource recommendation method provided in the embodiment of the present application can be used in a computer device. Optionally, the computer device is a terminal or a server. Optionally, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network) and big data and artificial intelligence platforms. Server. Optionally, the terminal is a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, intelligent voice interaction device, smart home appliance, vehicle-mounted terminal, aircraft, etc., but is not limited to this. The embodiments of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc.

[0190] In one possible implementation, the computer program involved in the embodiments of the present application may be deployed and executed on a computer device, or on multiple computer devices located at one location, or on multiple computer devices distributed at multiple locations and interconnected through a communication network. Multiple computer devices distributed at multiple locations and interconnected through a communication network can constitute a blockchain system.

[0191] In one possible implementation, the computer device in the embodiment of the present application is a node in a blockchain system, which can store object data, resource data, source domain interaction data, target domain interaction data, etc. in the blockchain. Thereafter, the node or the node corresponding to other devices in the blockchain can query the data stored in the blockchain by accessing the blockchain.

[0192] Figure 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application, see Figure 1 , the implementation environment includes: a terminal 101 and a server 102. The terminal 101 and the server 102 are connected via a wireless or wired network.

[0193] The terminal 101 is installed and runs a client 111, which may be a social application client, an online payment client, an online shopping client, a game client, a medical service client, a video client, etc. When the terminal 101 runs the client 111, the user interface of the client 111 is displayed on the screen of the terminal 101. The terminal 101 is a terminal used by the user 121.

[0194] Optionally, terminal 101 may generally refer to one of multiple terminals, including: smart phones, tablet computers, laptop computers, desktop computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, VR (Virtual Reality) devices, AR (Augmented Reality) devices, etc., but are not limited to these.

[0195] Those skilled in the art will appreciate that the number of the above terminals may be more or less. For example, the above terminal may be only one, or the above terminals may be 6, 8, or more. The embodiment of the present application does not limit the number of terminals and device types.

[0196] Figure 1 Only one terminal is shown, but in different embodiments, there are multiple other terminals 103 that can access the server 102. Optionally, there are one or more terminals 103 corresponding to the developer, and the client development and editing platform is installed on the terminal 103. The developer can edit and update the client on the terminal 103, and transmit the updated client installation package to the server 102 through a wired or wireless network. The terminal 101 can download the client installation package from the server 102 to update the client.

[0197] The terminal 101 and other terminals 103 are connected to the server 102 via a wired network or a wireless network.

[0198] The server 102 includes at least one of a single server, multiple servers, a cloud computing platform, and a virtualization center. The server 102 is used to provide background services for the client. Optionally, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work; or, the server 102 and the terminal 101 adopt a distributed computing architecture for collaborative computing.

[0199] In the embodiment of the present application, the terminal 101 sends a recommendation task for an object to the server 102. After receiving the recommendation task, the server 102 obtains the object data of the object and the resource data of the target resource to be recommended, and obtains the source domain interaction data and the target domain interaction data of the object, and the target resource belongs to the target domain. The server 102 determines the recommendation prediction result based on the object data of the object, the resource data of the target resource, the source domain interaction data and the target domain interaction data. If the recommendation prediction result indicates that if the target resource is recommended to the object, the object is more likely to interact with the target resource, then the server 102 returns the target resource to the terminal 101. The terminal 101 receives the target resource and recommends the target resource to the object.

[0200] It should be noted that the above implementation environment is only an example. The method provided in the embodiment of the present application can also be executed by the terminal 101 or the server 102 alone, or by other computer devices, and the embodiment of the present application does not limit this.

[0201] Figure 2 is a flowchart of a resource recommendation method provided in an embodiment of the present application. The embodiment of the present application is executed by a computer device, and the computer device is as follows Figure 1 Devices such as the terminal 101 or the server 102 are shown.

[0202] See also Figure 2 , the method comprising:

[0203] 201. A computer device obtains object data of an object and resource data of a target resource, where the target resource belongs to a target domain.

[0204] The computer device obtains object data of an object, which is an object of a resource to be recommended, for example, the object may be an account number, etc. The object data includes an object nickname, an object age, an object tag, and object preference information, etc.

[0205] The computer device obtains a target resource, which is a resource to be recommended to the object. In the embodiment of the present application, the resource may include a virtual resource, an item resource, an image resource, a video resource, a graphic resource, etc. For example, a virtual resource may be a voucher for exchanging an item.

[0206] In the embodiment of the present application, the target resource belongs to a target domain, and the target domain can be any domain. For example, the target domain can be an advertisement delivery domain, a video recommendation domain, an item recommendation domain, etc., which is not limited in the embodiment of the present application.

[0207] 202. The computer device obtains first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction matters, and the source domain interaction matters refer to the interaction between the object and the resources in the source domain. The target domain interaction data includes a plurality of target domain interaction matters, and the target domain interaction matters refer to the interaction between the object and the resources in the target domain. The source domain and the target domain belong to different fields.

[0208] In the embodiment of the present application, the source domain may be any domain different from the target domain, for example, the source domain may be a shopping domain, a video playback domain, etc.

[0209] The computer device obtains the first source domain interaction data of the object, and the first source domain interaction data is obtained based on the historical interaction between the object and at least one resource in the source domain. The first source domain interaction data includes multiple source domain interaction events, and a source domain interaction event refers to an event in which the object interacts with a resource in the source domain, that is, the source domain interaction event records the resources in the source domain that the object has interacted with.

[0210] The computer device obtains the target domain interaction data of the object, and the target domain interaction data is obtained based on the historical interaction between the object and at least one resource in the target domain. The target domain interaction data includes multiple target domain interaction events, and a target domain interaction event refers to an event in which the object interacts with a resource in the target domain, that is, the target domain interaction event records the resources in the target domain that the object has interacted with.

[0211] Optionally, the object and the resource can interact in a variety of ways. For example, if the resource is an item, the object and the resource interact means that the object purchases the item; for another example, if the resource is a virtual resource (such as a voucher), the object and the resource interact means that the object receives the virtual resource; for another example, if the resource is a video, the object and the resource interact means that the object plays the video; for another example, if the resource is an electronic book, the object and the resource interact means that the object reads the electronic book, etc.

[0212] 203. The computer device determines at least one source domain interaction item associated with the target domain in the first source domain interaction data, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data.

[0213] The first source domain interaction data includes multiple source domain interaction items, some of which are associated with the target domain, and some of which are irrelevant to the target domain. In the embodiment of the present application, the source domain interaction items are used as a reference to predict whether to recommend target resources in the target domain to the object. If the referenced source domain interaction items are irrelevant to the target domain, it will interfere with the prediction results and increase the processing burden.

[0214] Therefore, after determining the first source domain interaction data, at least one source domain interaction item associated with the target domain is screened out to form the second source domain interaction data, and the second source domain interaction data associated with the target domain is subsequently used to predict whether to recommend the target resource in the target domain to the object.

[0215] 204. The computer device determines a recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data. The recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0216] Since object data can reflect the characteristics of the object, resource data can reflect the characteristics of the target resource, the second source domain interaction data can reflect the object's interest in the resources in the source domain, and the target domain interaction data can reflect the object's interest in the resources in the target domain, the object data, resource data, second source domain interaction data, and target domain interaction data can be used to infer the object's interest in the target resource, and then determine the recommendation prediction result of the target resource. The recommendation prediction result is used to represent the object's interaction tendency with the target resource, that is, if the target resource is recommended to the object, how likely the object is to interact with the target resource. If the object is more likely to interact with the target resource, then the target resource can be recommended to the object later. If the object is less likely to interact with the target resource, then the target resource may not be recommended to the object later.

[0217] The method provided in the embodiment of the present application, when determining whether to recommend a target resource in the target domain for an object, needs to refer to the historical interaction of the object with other resources to infer the object's interaction tendency with the target resource. Therefore, the target domain interaction data of the object in the target domain is obtained. Considering that the target domain interaction data may have insufficient data, the source domain interaction data of the object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the source domain interaction data. Then, the target domain interaction data and the screened source domain interaction data are used as references to predict the interaction between the object and the target resource if the target resource is recommended to the object. On the one hand, since the source domain interaction data is introduced, the defect of insufficient target domain interaction data can be compensated, and the impact caused by the sparse target domain interaction data is alleviated. On the other hand, since only the source domain interaction data associated with the target domain is considered, the source domain interaction data unrelated to the target domain is discarded, so the interference of the source domain interaction data unrelated to the target domain on the prediction results can be reduced, thereby improving the accuracy of resource recommendation as a whole.

[0218] Based on the above embodiments, Figure 3 is a flowchart of another resource recommendation method provided by an embodiment of the present application. The embodiment of the present application is executed by a computer device, and the computer device is as follows Figure 1 The terminal 101 or the server 102 shown in FIG. Figure 3 , the method comprising:

[0219] 301. A computer device obtains object data of an object and resource data of a target resource, where the target resource belongs to a target domain.

[0220] In a possible implementation, the computer device obtains registration data of the object, resource data of resources that the object has interacted with, etc. as object data of the object. The registration data refers to data provided when registering the object, and may include a nickname of the object, an age of the object, etc.

[0221] In a possible implementation, the computer device obtains data of the target resource, such as the resource name, resource type, and entity to which the resource belongs, as resource data of the target resource.

[0222] 302. The computer device obtains first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction matters, and the source domain interaction matters refer to the interaction between the object and the resources in the source domain. The target domain interaction data includes a plurality of target domain interaction matters, and the target domain interaction matters refer to the interaction between the object and the resources in the target domain. The source domain and the target domain belong to different fields.

[0223] In one possible implementation, a computer device obtains multiple source domain interaction events that occurred to the object within a preset time period before a current time point, and uses the multiple source domain interaction events within the preset time period as first source domain interaction data.

[0224] Optionally, the preset duration is pre-set, such as 6 months, 3 months, etc.

[0225] Optionally, each source domain interaction event includes the source domain resources that the object has interacted with, the entity to which the source domain resources belong, and the interaction time point. The entity to which the source domain resources belong refers to the object that provides the source domain resources, for example, the entity may be a company or a manufacturer.

[0226] Exemplarily, in the case where the source domain is a shopping scenario, the source domain resources are purchased items, the source domain interactive items refer to shopping items, the computer device is associated with a target client, the target client provides a payment function, and the computer device stores the payment bill generated by the payment operation performed by the object in the target client. The computer device obtains multiple payment bills of the object generated within a preset time period between the current time point, and for each payment bill, the computer device generates a source domain interactive item based on the payment bill. The source domain interactive item includes the items purchased in the payment bill, the entity to which the items belong, and the payment time point. The purchased items are also the above-mentioned source domain resources, and the payment time point is also the above-mentioned interaction time point.

[0227] In one possible implementation, the computer device obtains multiple target domain interaction events that occurred to the object within a preset time period before the current time point, and uses the multiple target domain interaction events within the preset time period as first target domain interaction data.

[0228] Optionally, each target domain interaction event includes the target domain resource that the object has interacted with, the entity to which the target domain resource belongs, and the interaction time point. The entity to which the target domain resource belongs refers to the object that provides the target domain resource, for example, the entity may be a company or a manufacturer.

[0229] Exemplarily, in the case where the target domain is an advertising delivery scenario, the target domain resources are virtual resources contained in the delivered advertisements, and the virtual resources are used to exchange items in the source domain. The target domain interactive event refers to a virtual resource collection event, and the computer device is associated with a target client. The target client provides an advertising delivery function, and advertisements containing virtual resources can be delivered in the target client. The computer device stores a collection record generated by the virtual resource collection operation performed by the object in the target client. The computer device obtains multiple collection records of the object generated within a preset time period between the current time point. For each collection record, the computer device generates a target domain interactive event based on the collection record. The target domain interactive event includes the virtual resource collected in the collection record, the entity to which the virtual resource belongs, and the collection time point. The purchased virtual resource is also the above-mentioned target domain resource, and the collection time point is also the above-mentioned interaction time point.

[0230] 303. The computer device determines at least one source domain interaction item associated with the target domain in the first source domain interaction data, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data.

[0231] The computer device stores the interaction data of multiple objects, each of which has its own source domain interaction items and target domain interaction items. The amount of data of the source domain interaction items of each object is huge. Figure 4 As shown, the total number of source domain interaction items and target domain interaction items of multiple objects are counted. The maximum total number of source domain interaction items is about 5000, the median total number of source domain interaction items is about 1000, and the maximum total number of target domain interaction items is far less than 1000.

[0232] Therefore, the number of multiple source domain interaction items in the first source domain interaction data acquired by the computer device is also large. However, among the multiple source domain interaction items, some source domain interaction items are associated with the target domain, and some source domain interaction items are not related to the target domain. In an embodiment of the present application, it is necessary to use the source domain interaction items as a reference to predict whether to recommend the target resources in the target domain to the object. If the referenced source domain interaction items are not related to the target domain, it will interfere with the prediction results, and due to the large amount of data, it will also increase the processing burden.

[0233] Based on this, after determining the first source domain interaction data, the embodiment of the present application selects at least one source domain interaction item associated with the target domain to form the second source domain interaction data, so as to ensure that all source domain interaction items in the second source domain interaction data are related to the target domain, and then use the second source domain interaction data to predict whether to recommend the target resource in the target domain to the object. On the one hand, it can ensure that the referenced source domain interaction data is strongly related to the target domain, and on the other hand, it can effectively reduce the amount of data to be processed and reduce the processing burden.

[0234] In one possible implementation, a computer device determines multiple target entities, where a target entity refers to an entity to which resources in a target domain belong, and an entity refers to an object that provides resources; for any source domain interaction item in the first source domain interaction data, if the entity to which the resources in the source domain interaction item belong is any target entity, it is determined that the source domain interaction item is associated with the target domain.

[0235] The resources in the target domain have their own entities, and the entities to which the resources in the target domain belong are determined as target entities. The resources in the source domain have their own entities, and for any source domain interaction event, if the entity to which the interactive resources in the source domain interaction event belong is the same as any target entity, then the source domain interaction event can be considered to be associated with the target domain.

[0236] In an embodiment of the present application, taking into account that users may have similar interaction tendencies towards different resources provided by the same entity, if the entity to which the resources in the source domain interaction matter belong is the same as the entity to which the resources in the target domain belong, then the source domain interaction matter can be considered to be associated with the target domain, and the source domain interaction matter can be used as a reference to predict the user's interaction tendency towards the resources provided by the same entity in the target domain, which is conducive to ensuring the accuracy of cross-domain mining of user interests, thereby improving the accuracy of resource recommendations.

[0237] Moreover, searching for source domain interaction items associated with the target domain based on the condition of entity identity can ensure that the source domain interaction items found have a strong correlation with the target domain, which is conducive to improving the accuracy of searching for source domain interaction items associated with the target domain.

[0238] In another possible implementation, a computer device determines multiple target entities, where a target entity refers to an entity to which resources in a target domain belong, and an entity refers to an object that provides resources; for any source domain interaction event in the first source domain interaction data, if the similarity between the entity to which the resources in the source domain interaction event belong and any target entity is greater than a similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0239] For any source domain interaction event, if the similarity between the entity to which the interactive source domain resource belongs in the source domain interaction event and any target entity is greater than a similarity threshold, the source domain interaction event can be considered to be associated with the target domain.

[0240] Optionally, the computer device may use a large language model to determine the similarity between an entity to which a resource in a source domain interactive matter belongs and any target entity.

[0241] Optionally, the computer device obtains target entity features of each target entity; performs feature extraction on entities to which resources in source domain interactive events belong to obtain source domain entity features; if the similarity between the source domain entity features and any target entity features is greater than a similarity threshold, it is determined that the source domain interactive event is associated with the target domain.

[0242] Exemplarily, the computer device obtains the entity name of the target entity, inputs the entity name of the target entity into the text semantic extraction model, and obtains the target entity feature output by the text semantic extraction model. The computer device obtains the entity name of the entity to which the source domain resource belongs, inputs the entity name of the entity to which the source domain resource belongs into the text semantic extraction model, and obtains the source domain entity feature output by the text semantic extraction model.

[0243] The text semantic extraction model is used to extract semantic features, and the text semantic extraction model may be a deep learning model, a machine learning model, or a large language model, etc. For example, the text semantic extraction model may be a BGE (BAAI General Embedding, text semantic vector model).

[0244] Exemplarily, the computer device determines the similarity between the source domain entity feature and each target entity feature, and obtains the maximum similarity among the determined multiple similarities. If the maximum similarity is greater than a similarity threshold, the source domain interaction item is determined to be associated with the target domain.

[0245] The similarity may be cosine similarity, etc. The similarity threshold may be a preset threshold, for example, the similarity threshold may be equal to 0.8 or 0.85, etc.

[0246] In an embodiment of the present application, taking into account that users may have similar interaction tendencies towards different resources provided by similar entities, if the entity to which the resources in the source domain interaction matter belong is similar to the entity to which the resources in the target domain belong, then the source domain interaction matter can be considered to be associated with the target domain, and the source domain interaction matter can be used as a reference to predict the user's interaction tendency towards resources provided by similar entities in the target domain, which is conducive to ensuring the accuracy of cross-domain mining of user interests, thereby improving the accuracy of resource recommendations.

[0247] Moreover, searching for source domain interaction items related to the target domain based on entity similarity can cover more source domain interaction items related to the target domain, which is conducive to ensuring the data volume of the source domain interaction items found and avoiding the problems of data sparsity and insufficient mining.

[0248] Furthermore, determining the similarity between entities based on the similarity between the features of the entities is conducive to ensuring the accuracy of the determined similarity, and further ensuring the accuracy of the source domain interaction items associated with the target domain that are found.

[0249] In a possible implementation, if the length of the second source domain interaction data is less than a preset length, the second source domain interaction data is padded so that the length of the padded second source domain interaction data is equal to the preset length. Optionally, the padding value may be 0. If the length of the second source domain interaction data is greater than a preset length, the second source domain interaction data is truncated so that the length of the truncated second source domain interaction data is equal to the preset length. Optionally, truncating the second source domain interaction data means deleting source domain interaction items with relatively earlier interaction time points in the order of interaction time points from earliest to latest.

[0250] The length of the second source domain interaction data is the number of source domain interaction items in the second source domain interaction data.

[0251] In a possible implementation, if the length of the target domain interaction data is less than a preset length, the target domain interaction data is padded so that the length of the padded target domain interaction data is equal to the preset length. Optionally, the padding value may be 0. If the length of the target domain interaction data is greater than the preset length, the target domain interaction data is truncated so that the length of the truncated target domain interaction data is equal to the preset length. Optionally, truncating the target domain interaction data means deleting target domain interaction items with relatively earlier interaction time points in the order of interaction time points from earliest to latest.

[0252] The length of the target domain interaction data is the number of target domain interaction items in the target domain interaction data.

[0253] 304. The computer device extracts features from the object data to obtain object features.

[0254] After acquiring the object data, the computer device extracts features from the object data to obtain object features.

[0255] Optionally, the computer device may utilize a feature extraction sub-model to perform the feature extraction step. The computer device inputs the object data into the feature extraction sub-model, and the feature extraction sub-model extracts features from the object data and outputs object features.

[0256] Optionally, the object feature may be a feature vector or a feature matrix, etc.

[0257] 305. The computer device extracts features from the resource data to obtain resource features.

[0258] After acquiring the resource data, the computer device extracts features from the resource data to obtain resource features.

[0259] Optionally, the computer device may utilize a feature extraction sub-model to perform a feature extraction step. The computer device inputs resource data into the feature extraction sub-model, and the feature extraction sub-model extracts features from the resource data and outputs resource features.

[0260] Optionally, the resource feature may be a feature vector or a feature matrix, etc.

[0261] 306. The computer device extracts features from the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features.

[0262] After acquiring the second source domain interaction data and the target domain interaction data, the computer device extracts features from the second source domain interaction data and the target domain interaction data to obtain source domain interaction features of the second source domain interaction data and target domain interaction features of the target domain interaction data.

[0263] In a possible implementation, the second source domain interaction data includes multiple source domain interaction items, each of which includes a source domain resource, an entity to which the source domain resource belongs, and an interaction time point. The source domain interaction features include features of multiple source domain interaction items, each of which includes source domain resource features, source domain entity features, and source domain interaction time features.

[0264] In a possible implementation, the target domain interaction data includes multiple target domain interaction items, each of which includes a target domain resource, an entity to which the target domain resource belongs, and an interaction time point. The target domain interaction features include features of multiple target domain interaction items, each of which includes a target domain resource feature, a target domain entity feature, and a target domain interaction time feature.

[0265] In one possible implementation, the computer device uses an attention mechanism to process the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features, and step 306 includes the following steps 3061-3065.

[0266] 3061. Perform feature extraction on the second source domain interaction data to obtain initial source domain interaction features, and perform feature extraction on the target domain interaction data to obtain initial target domain interaction features.

[0267] The computer device extracts features from the second source domain interaction data and the target domain interaction data, respectively, to obtain initial source domain interaction features and initial target domain interaction features, respectively. The initial source domain interaction features represent features of the second source domain interaction data itself without being affected by the target domain interaction data, and the initial target domain interaction features represent features of the target domain interaction data itself without being affected by the second source domain interaction data.

[0268] However, although the source domain interaction items in the second source domain interaction data are all associated with the target domain, the degree of association between each source domain interaction item and the target domain is different. In order to improve accuracy, the influence of the source domain interaction features with a relatively higher degree of association with the target domain on the recommendation prediction results should be enhanced, and the influence of the source domain interaction features with a relatively lower degree of association with the target domain on the recommendation prediction results should be reduced. In addition, there may be noise in the target domain interaction data. For example, when a target domain interaction item is completely unrelated to the source domain, the target domain interaction item may be caused by a user's accidental touch, so the influence of the target domain interaction features of such target domain interaction items on the recommendation prediction results should be reduced. Based on this, in the following steps 3062-3064, the embodiment of the present application uses an attention mechanism to process the initial source domain interaction features and the initial target domain interaction features to explore the importance of each part of the initial source domain interaction features and the initial target domain interaction features, so as to enhance the more important parts of the initial source domain interaction features and the initial target domain interaction features.

[0269] 3062. Fuse the initial source domain interaction feature and the initial target domain interaction feature to obtain a first fused interaction feature.

[0270] Optionally, the computer device concatenates the initial source domain interaction feature and the initial target domain interaction feature to obtain a first fused interaction feature.

[0271] Exemplarily, the computer device uses the following formula to determine the first fusion interaction feature.

[0272] HC=[HS gs ||HT]

[0273] Among them, HC represents the first fusion interaction feature, HS gs represents the initial source domain interaction feature, HT represents the initial target domain interaction feature, and || represents the concatenation operation.

[0274] Optionally, the computer device performs weighted summation of the initial source domain interaction feature and the initial target domain interaction feature to obtain a first fused interaction feature.

[0275] 3063. Multiply the first fused interaction feature by the first mapping matrix to obtain a first query feature, a first key feature, and a first value feature.

[0276] The key feature (key), value feature (value) and query feature (query) in the embodiment of the present application belong to different feature spaces respectively. The value feature represents the feature of the first fused interactive feature itself, and the query feature and key feature are used to determine the importance of the features corresponding to each interactive item in the first fused interactive feature.

[0277] The first mapping matrix is ​​a preset parameter matrix, and the first mapping matrix is ​​used to perform spatial transformation on the first fusion interaction feature.

[0278] Optionally, the first mapping matrix is ​​a mapping matrix of three dimensions, including a mapping matrix of a query dimension, a mapping matrix of a key dimension, and a mapping matrix of a value dimension. The computer device multiplies the mapping matrices of the three dimensions by the first fusion interaction feature to obtain a first query feature, a first key feature, and a first value feature, respectively.

[0279] Optionally, the computer device uses the following formula to determine the first query feature, the first key feature and the first value feature.

[0280]

[0281] Among them, Q1 represents the first query feature, K1 represents the first key feature, V1 represents the first value feature. HC represents the first fusion interaction feature, represents the mapping matrix of query dimensions, represents the mapping matrix of the key dimension, A mapping matrix representing the value dimension.

[0282] 3064. Fuse the first query feature, the first key feature, and the first value feature to obtain a second fused interactive feature.

[0283] Optionally, the computer device normalizes the product of the transposition of the first query feature and the first key feature and the scaling factor to obtain a first normalized feature, and determines the product of the first normalized feature and the first value feature as the second fused interactive feature.

[0284] Exemplarily, the computer device uses the following formula to fuse the first query feature, the first key feature, and the first value feature.

[0285]

[0286] Among them, Attention(Q1,K1,V1) represents the second fusion interaction feature, Q1 represents the first query feature, K1 represents the first key feature, and V1 represents the first value feature. T represents the transpose operation of the matrix. represents the scaling factor and Softmax(·) represents the normalization operation.

[0287] 3065. Based on the second fused interactive features, determine the source domain interactive features and the target domain interactive features.

[0288] After determining the second fused interactive feature, the computer device splits the second fused interactive feature into a source domain interactive feature and a target domain interactive feature. Optionally, the first fused interactive feature is obtained by splicing the initial source domain interactive feature and the initial target domain interactive feature, then the second fused interactive feature is split again according to the splicing method, and the two split interactive features are used as the source domain interactive feature and the target domain interactive feature that are finally used.

[0289] For example, the first fused interactive feature is obtained by concatenating the initial source domain interactive feature and the initial target domain interactive feature, the length of the initial source domain interactive feature is m, the length of the initial target domain interactive feature is n, and the first fused interactive feature and the second fused interactive feature are both m+n. Then the computer device uses the 1st to mth items in the second fused interactive feature as the source domain interactive features, and the m+1st to m+nth items as the target domain interactive features.

[0290] Optionally, the process of determining the source domain interaction feature in step 3065 includes: splitting out the intermediate source domain interaction feature from the second fused interaction feature; multiplying the resource feature with the second mapping matrix to obtain the second query feature; multiplying the intermediate source domain interaction feature with the second mapping matrix to obtain the second key feature and the second value feature; and fusing the second query feature, the second key feature and the second value feature to obtain the source domain interaction feature.

[0291] In an embodiment of the present application, the computer device uses the interactive features belonging to the source domain separated from the second fused interactive features as intermediate source domain interactive features, and uses the attention mechanism to process the intermediate source domain interactive features and resource features to obtain the source domain interactive features that are finally used.

[0292] The second mapping matrix is ​​a preset parameter matrix, and the second mapping matrix is ​​used to perform spatial transformation on resource features and intermediate source domain interaction features.

[0293] Exemplarily, the second mapping matrix is ​​a mapping matrix of three dimensions, including a mapping matrix of a query dimension, a mapping matrix of a key dimension, and a mapping matrix of a value dimension. The computer device multiplies the mapping matrix of the query dimension with the resource feature to obtain a second query feature, and multiplies the mapping matrix of the key dimension and the mapping matrix of the value dimension with the intermediate source domain interaction feature to obtain a second key feature and a second value feature, respectively.

[0294] For example, the computer device uses the following formula to determine the second query feature, the second key feature, and the second value feature.

[0295]

[0296] Wherein, Q2 represents the second query feature, K2 represents the second key feature, and V2 represents the second value feature. t Indicates resource characteristics, HS sa Represents the intermediate source domain interaction characteristics. represents the mapping matrix of query dimensions, represents the mapping matrix of the key dimension, A mapping matrix representing the value dimension.

[0297] Exemplarily, the computer device normalizes the product of the transposition of the second query feature and the second key feature and the scaling factor to obtain a second normalized feature, and determines the product of the second normalized feature and the second value feature as the source domain interaction feature.

[0298] For example, the computer device uses the following formula to fuse the second query feature, the second key feature, and the second value feature.

[0299]

[0300] Among them, Attention(Q2,K2,V2) represents the source domain interaction feature, Q2 represents the second query feature, K2 represents the second key feature, and V2 represents the second value feature. T represents the transpose operation of the matrix. represents the scaling factor and Softmax(·) represents the normalization operation.

[0301] Optionally, the process of determining the target domain interaction feature in step 3065 includes: splitting out the intermediate target domain interaction feature from the second fused interaction feature; multiplying the resource feature with the third mapping matrix to obtain the third query feature; multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain the third key feature and the third value feature; and fusing the third query feature, the third key feature and the third value feature to obtain the target domain interaction feature.

[0302] In an embodiment of the present application, the computer device uses the interactive features belonging to the target domain separated from the third fused interactive features as intermediate target domain interactive features, and uses the attention mechanism to process the intermediate target domain interactive features and resource features to obtain the target domain interactive features that are finally used.

[0303] The third mapping matrix is ​​a preset parameter matrix, and the third mapping matrix is ​​used to perform spatial transformation on resource features and intermediate target domain interaction features.

[0304] Exemplarily, the third mapping matrix is ​​a mapping matrix of three dimensions, including a mapping matrix of a query dimension, a mapping matrix of a key dimension, and a mapping matrix of a value dimension. The computer device multiplies the mapping matrix of the query dimension with the resource feature to obtain a third query feature, and multiplies the mapping matrix of the key dimension and the mapping matrix of the value dimension with the intermediate target domain interaction feature to obtain a third key feature and a third value feature, respectively.

[0305] For example, the computer device uses the following formula to determine the third query feature, the third key feature, and the third value feature.

[0306]

[0307] Among them, Q3 represents the third query feature, K3 represents the third key feature, and V3 represents the third value feature. t Indicates resource characteristics, HT sa Represents the intermediate target domain interaction features. represents the mapping matrix of query dimensions, represents the mapping matrix of the key dimension, A mapping matrix representing the value dimension.

[0308] Exemplarily, the computer device normalizes the product of the transpose of the third query feature and the third key feature and the scaling factor to obtain a third normalized feature, and determines the product of the third normalized feature and the third value feature as the target domain interaction feature.

[0309] For example, the computer device uses the following formula to fuse the third query feature, the third key feature, and the third value feature.

[0310]

[0311] Among them, Attention(Q3,K3,V3) represents the target domain interaction feature, Q3 represents the third query feature, K3 represents the third key feature, and V3 represents the third value feature. T represents the transpose operation of the matrix. represents the scaling factor and Softmax(·) represents the normalization operation.

[0312] In the embodiment of the present application, the attention mechanism is used to process the source domain interaction features and the target domain interaction features. On the one hand, the correlation between the source domain interaction features and the target domain interaction features can be used to weight the source domain interaction features, which is conducive to enhancing the source domain interaction features with higher correlation with the target domain and weakening the source domain interaction features with lower correlation with the target domain. On the other hand, the correlation between the target domain interaction features and the source domain interaction features can be used to weight the target domain interaction features, so as to adjust the target domain interaction features using the source domain interaction features, which is conducive to weakening the characteristics of sparse interactive behaviors in the target domain, and to a certain extent, can reduce the interference of the interactive behaviors that are accidentally touched in the target domain.

[0313] In addition, by using the attention mechanism to process the source domain interaction features and the resource features of the target resources, the correlation between the source domain interaction features and the resource features can be used to weight the source domain interaction features, which is conducive to enhancing the source domain interaction features that are more correlated with the target resources, thereby mining the source domain interaction features that are strongly correlated with the target resources, so that the source domain interaction features referenced when recommending target resources are concentrated on the source domain interaction features related to the target resources to be recommended, reducing the interference of irrelevant source domain interaction features, and thus improving the accuracy of the recommendation prediction results of the target resources.

[0314] In addition, by using the attention mechanism to process the target domain interaction features and the resource features of the target resources, the correlation between the target domain interaction features and the resource features can be used to weight the target domain interaction features, which is conducive to enhancing the target domain interaction features that are more correlated with the target resources, thereby mining the target domain interaction features that are strongly correlated with the target resources, so that the target domain interaction features referenced when recommending target resources are concentrated on the target domain interaction features related to the target resources to be recommended, reducing the interference of irrelevant target domain interaction features, and thus improving the accuracy of the recommendation prediction results of the target resources.

[0315] 307. The computer device determines a recommendation prediction result of the target resource based on the object characteristics, resource characteristics, source domain interaction characteristics and target domain interaction characteristics. The recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0316] After determining the object features, resource features, source domain interaction features and target domain interaction features, the computer device can use the object features, resource features, source domain interaction features and target domain interaction features to infer the object's interest tendency towards the target resource, and then determine the recommendation prediction result of the target resource. The recommendation prediction result is used to represent the object's interaction tendency towards the target resource.

[0317] In a possible implementation, the recommendation prediction result is used to represent the probability of interaction between the object and the target resource if the target resource is recommended to the object. For example, the recommendation prediction result may be CTR (Click-Through-Rate).

[0318] In an embodiment of the present application, the interaction between an object and a target resource is predicted based on the characteristics of the object data, the characteristics of the resource data, the characteristics of the source domain interaction data, and the characteristics of the target domain interaction data. Through the potential characteristics of these data, the interest tendencies of the object and the degree of match between the interest tendencies of the object and the target resource can be fully explored, thereby inferring the object's interaction tendency with the target resource, which is conducive to ensuring the accuracy of the recommendation prediction results.

[0319] 308. If the recommendation prediction result meets the recommendation condition, the computer device recommends the target resource to the object.

[0320] After determining the recommendation prediction result, the computer device determines whether the recommendation prediction result meets the recommendation condition, and if so, recommends the target resource to the object. If not, there is no need to recommend the target resource to the object.

[0321] In a possible implementation, the recommendation prediction result is used to characterize the probability of interaction between the object and the target resource if the target resource is recommended to the object. The recommendation condition is that the interaction probability of the target resource is greater than a probability threshold, or the recommendation condition is that among the interaction probabilities of multiple candidate recommended resources, the interaction probability of the target resource ranks first by a preset number in order of magnitude.

[0322] Figure 5 is a schematic diagram of a resource recommendation method provided in an embodiment of the present application, such as Figure 5 As shown in the figure, the user views the product interface and requests the resource recommendation service in the product interface. The server obtains object data, resource data, source domain interaction data, and target domain interaction data from online storage. Using the resource recommendation model, the recommendation prediction result of the target resource is determined based on the acquired data. If the recommendation conditions are met, the target resource is recommended to the user. In addition, after the user interacts with the target resource, a log is generated. The server stores the log in the log system. The target domain interaction items of the object can be extracted from the log later for the next recommendation task.

[0323] The method provided in the embodiment of the present application, when determining whether to recommend a target resource in the target domain for an object, needs to refer to the historical interaction of the object with other resources to infer the object's interaction tendency with the target resource. Therefore, the target domain interaction data of the object in the target domain is obtained. Considering that the target domain interaction data may have insufficient data, the source domain interaction data of the object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the source domain interaction data. Then, the target domain interaction data and the screened source domain interaction data are used as references to predict the interaction between the object and the target resource if the target resource is recommended to the object. On the one hand, since the source domain interaction data is introduced, the defect of insufficient target domain interaction data can be compensated, and the impact caused by the sparse target domain interaction data is alleviated. On the other hand, since only the source domain interaction data associated with the target domain is considered, the source domain interaction data unrelated to the target domain is discarded, so the interference of the source domain interaction data unrelated to the target domain on the prediction results can be reduced, thereby improving the accuracy of resource recommendation as a whole.

[0324] Moreover, when the recommendation prediction result meets the recommendation conditions, it indicates that the object is more likely to interact with the target resource. Therefore, the target resource is recommended to the object, thereby realizing personalized resource recommendation to the object, which is conducive to improving the conversion rate of resource recommendation and enhancing the recommendation effect.

[0325] Based on the above embodiment, the above steps 304 to 307 can be implemented using a resource recommendation model. The resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model. The detailed process can be found in the following Figure 6 Embodiment of the invention.

[0326] Figure 6 is a flowchart of another resource recommendation method provided by an embodiment of the present application. The embodiment of the present application is executed by a computer device, and the computer device is as follows Figure 1 The terminal 101 or the server 102 shown in FIG. Figure 6 , the method comprising:

[0327] 601. A computer device obtains object data of an object and resource data of a target resource, where the target resource belongs to a target domain.

[0328] 602. The computer device obtains first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction matters, and the source domain interaction matters refer to the interaction between the object and the resources in the source domain. The target domain interaction data includes a plurality of target domain interaction matters, and the target domain interaction matters refer to the interaction between the object and the resources in the target domain. The source domain and the target domain belong to different fields.

[0329] 603. The computer device determines at least one source domain interaction item associated with the target domain in the first source domain interaction data, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data.

[0330] Among them, the above steps 601 to 603 are the same as the above steps 301 to 303, and will not be repeated here.

[0331] 604. The computer device extracts features from the object data through the feature extraction sub-model in the resource recommendation model to obtain object features.

[0332] The feature extraction sub-model in the resource recommendation model is used to extract features. The computer device inputs the object data into the feature extraction sub-model, and the feature extraction sub-model extracts features from the object data and outputs object features.

[0333] The process of extracting features from the object data by the feature extraction sub-model is the same as the above step 304 and will not be described in detail here.

[0334] 605. The computer device extracts features from the resource data through the feature extraction sub-model to obtain resource features.

[0335] The computer device inputs the resource data into the feature extraction sub-model, and the feature extraction sub-model extracts features from the resource data and outputs resource features.

[0336] The process of extracting features from resource data by the feature extraction sub-model is the same as the above step 305 and will not be described in detail here.

[0337] 606. The computer device extracts features from the second source domain interaction data and the target domain interaction data through a feature extraction sub-model to obtain source domain interaction features and target domain interaction features.

[0338] The computer device inputs the second source domain interaction data and the target domain interaction data into the feature extraction sub-model, and the feature extraction sub-model extracts features from the second source domain interaction data and the target domain interaction data, and outputs source domain interaction features and target domain interaction features.

[0339] In a possible implementation, the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, and the first attention network includes a first mapping matrix. Step 606 includes the following steps 6061 to 6065.

[0340] 6061. Perform feature extraction on the second source domain interaction data through a feature extraction network to obtain initial source domain interaction features.

[0341] 6062. Through the feature extraction network, feature extraction is performed on the target domain interaction data to obtain initial target domain interaction features.

[0342] 6063. The initial source domain interaction feature and the initial target domain interaction feature are fused through a feature fusion network to obtain a first fused interaction feature.

[0343] 6064. Multiply the first fused interactive feature by the first mapping matrix through the first attention network to obtain a first query feature, a first key feature and a first value feature, and fuse the first query feature, the first key feature and the first value feature to obtain a second fused interactive feature.

[0344] 6065. Based on the second fused interaction feature, determine the source domain interaction feature and the target domain interaction feature.

[0345] Optionally, the feature extraction sub-model also includes a second attention network, and the second attention network includes a second mapping matrix.

[0346] The process of determining the source domain interaction feature in step 6065 includes: splitting the intermediate source domain interaction feature from the second fused interaction feature; multiplying the resource feature with the second mapping matrix through the second attention network to obtain the second query feature, and multiplying the intermediate source domain interaction feature with the second mapping matrix to obtain the second key feature and the second value feature; and fusing the second query feature, the second key feature and the second value feature through the second attention network to obtain the source domain interaction feature.

[0347] Optionally, the feature extraction sub-model also includes a third attention network, and the third attention network includes a third mapping matrix.

[0348] The process of determining the target domain interaction feature in step 6065 includes: splitting the intermediate target domain interaction feature from the second fused interaction feature; multiplying the resource feature with the third mapping matrix through the third attention network to obtain the third query feature, and multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain the third key feature and the third value feature; and fusing the third query feature, the third key feature and the third value feature through the third attention network to obtain the target domain interaction feature.

[0349] In the embodiment of the present application, an attention network is used to process the source domain interaction features and the target domain interaction features. On the one hand, the correlation between the source domain interaction features and the target domain interaction features can be used to weight the source domain interaction features, which is conducive to enhancing the source domain interaction features with higher correlation with the target domain and weakening the source domain interaction features with lower correlation with the target domain. On the other hand, the correlation between the target domain interaction features and the source domain interaction features can be used to weight the target domain interaction features, so as to adjust the target domain interaction features using the source domain interaction features, which is conducive to weakening the characteristics of sparse interactive behaviors in the target domain, and to a certain extent, can reduce the interference of the interactive behaviors that are accidentally touched in the target domain.

[0350] In addition, by using the attention network to process the source domain interaction features and the resource features of the target resources, the correlation between the source domain interaction features and the resource features can be used to weight the source domain interaction features, which is conducive to enhancing the source domain interaction features that are more correlated with the target resources, thereby mining the source domain interaction features that are strongly correlated with the target resources, so that the source domain interaction features referenced when recommending target resources are concentrated on the source domain interaction features related to the target resources to be recommended, reducing the interference of irrelevant source domain interaction features, and thus improving the accuracy of the recommendation prediction results of the target resources.

[0351] In addition, by using the attention network to process the target domain interaction features and the resource features of the target resources, the correlation between the target domain interaction features and the resource features can be used to weight the target domain interaction features, which is conducive to enhancing the target domain interaction features that are more correlated with the target resources, thereby mining the target domain interaction features that are strongly correlated with the target resources, so that the target domain interaction features referenced when recommending target resources are concentrated on the target domain interaction features related to the target resources to be recommended, reducing the interference of irrelevant target domain interaction features, and thus improving the accuracy of the recommendation prediction results of the target resources.

[0352] It should be noted that, when the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, in the above steps 604 and 605, it is only necessary to extract object features and resource features through the feature extraction network.

[0353] 607. The computer device determines the recommendation prediction result of the target resource based on the object characteristics, resource characteristics, source domain interaction characteristics and target domain interaction characteristics through the recommendation prediction sub-model in the resource recommendation model. The recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0354] The recommendation prediction submodel in the resource recommendation model is used to determine the recommendation prediction result. The computer device inputs the object features, resource features, source domain interaction features and target domain interaction features into the recommendation prediction submodel, and the recommendation prediction submodel outputs the recommendation prediction result. Exemplarily, the recommendation prediction submodel can be an MLP (Multilayer Perceptron).

[0355] In a possible implementation, the computer device pools the source domain interaction features and the target domain interaction features to obtain the pooled source domain interaction features and the pooled target domain interaction features. The object features, resource features, the pooled source domain interaction features and the pooled target domain interaction features are spliced ​​to obtain the spliced ​​features, and the spliced ​​features are input into the recommendation prediction sub-model, which outputs the recommendation prediction results.

[0356] Optionally, the source domain interaction feature includes source domain interaction item features of multiple source domain interaction items, and the target domain interaction feature includes target domain interaction item features of multiple target domain interaction items. Then the computer device averagely pools the multiple source domain interaction item features to obtain the pooled source domain interaction feature. The computer device averagely pools the multiple target domain interaction item features to obtain the pooled target domain interaction feature.

[0357] Exemplarily, the computer device uses the following formula to pool the source domain interaction features and the target domain interaction features.

[0358]

[0359] Among them, rs represents the source domain interaction feature after pooling, hs represents a source domain interaction event feature in the source domain interaction feature, and m represents the length of the source domain interaction feature, that is, the number of source domain interaction event features. rt represents the target domain interaction feature after pooling, ht represents a target domain interaction event feature in the target domain interaction feature, and n represents the length of the target domain interaction feature, that is, the number of target domain interaction event features.

[0360] Exemplarily, the computer device uses the following formula to determine the recommended prediction result.

[0361]

[0362] in, represents the recommendation prediction result, x represents the object feature, q t represents resource features, rs represents the source domain interaction features after pooling, rt represents the target domain interaction features after pooling, || represents the concatenation operation, and Model(·) represents the recommendation prediction sub-model.

[0363] 608. If the recommendation prediction result meets the recommendation condition, the computer device recommends the target resource to the object.

[0364] The method provided in the embodiment of the present application, when determining whether to recommend a target resource in the target domain for an object, needs to refer to the historical interaction of the object with other resources to infer the object's interaction tendency with the target resource. Therefore, the target domain interaction data of the object in the target domain is obtained. Considering that the target domain interaction data may have insufficient data, the source domain interaction data of the object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the source domain interaction data. Then, the target domain interaction data and the screened source domain interaction data are used as references to predict the interaction between the object and the target resource if the target resource is recommended to the object. On the one hand, since the source domain interaction data is introduced, the defect of insufficient target domain interaction data can be compensated, and the impact caused by the sparse target domain interaction data is alleviated. On the other hand, since only the source domain interaction data associated with the target domain is considered, the source domain interaction data unrelated to the target domain is discarded, so the interference of the source domain interaction data unrelated to the target domain on the prediction results can be reduced, thereby improving the accuracy of resource recommendation as a whole.

[0365] Furthermore, using a resource recommendation model to execute a process of determining a recommendation prediction result based on object data, resource data, source domain interaction data, and target domain interaction data is beneficial to improving processing efficiency.

[0366] The embodiment of the present application also provides a method for training a resource recommendation model. Figure 7 is a flowchart of a method for training a resource recommendation model provided in an embodiment of the present application. The embodiment of the present application is executed by a computer device, and the computer device is as follows Figure 1 The terminal 101 or the server 102 shown in FIG. Figure 7 , the method comprising:

[0367] 701. A computer device obtains sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object.

[0368] Among them, the sample resource is a resource that has been recommended to the sample object, the sample object data is the same as the above object data, and the sample resource data is the same as the above resource data, which will not be repeated here.

[0369] The sample label may be the actual interaction probability between the sample object and the sample resource after the sample resource is recommended to the sample object. For example, if the sample object interacts with the sample resource, the sample label may be set to 1, and if the sample object does not interact with the sample resource, the sample label may be set to 0.

[0370] 702. The computer device obtains first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to the interaction between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to the interaction between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains.

[0371] 703. The computer device determines at least one sample source domain interaction item associated with the target domain in the first sample source domain interaction data, and the at least one sample source domain interaction item associated with the target domain constitutes the second sample source domain interaction data.

[0372] 704. The computer device determines a recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data and the target domain sample interaction data through a resource recommendation model. The recommendation prediction result is used to characterize the predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object.

[0373] The process of step 702 to step 704 is the same as the process of step 602 to step 607, which will not be repeated here.

[0374] 705. The computer device trains the resource recommendation model based on the recommendation prediction results and the sample labels.

[0375] The sample label can reflect the actual interaction between the sample object and the sample resource, and the recommendation prediction result is the prediction result obtained based on the resource recommendation model, which can reflect the predicted interaction between the sample object and the sample resource. The smaller the difference between the recommendation prediction result and the sample label, the more accurate the recommendation prediction result, that is, the more accurate the resource recommendation model. Therefore, the computer device trains the resource recommendation model based on the recommendation prediction result and the sample label. The training goal is to reduce the difference between the recommendation prediction result and the sample label, thereby improving the accuracy of the resource recommendation model.

[0376] The method provided in the embodiment of the present application, when training the resource recommendation model, for the recommendation task of the target domain, obtains the sample target domain interaction data of the sample object in the target domain. Considering that the sample target domain interaction data may have insufficient data, the sample source domain interaction data of the sample object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the sample source domain interaction data, and then the sample target domain interaction data and the screened sample source domain interaction data are used as references to use the resource recommendation model to predict the recommendation prediction result of recommending the sample resource to the sample object, and then the resource recommendation model is trained using the recommendation prediction result and the sample label. On the one hand, due to the introduction of the sample source domain interaction data, the defect of insufficient sample target domain interaction data can be compensated, the impact caused by the sparse sample target domain interaction data is alleviated, and it is beneficial to improve the accuracy of the resource recommendation model. On the other hand, since only the sample source domain interaction data associated with the target domain is considered, the sample source domain interaction data irrelevant to the target domain is discarded, so the interference caused by the sample source domain interaction data irrelevant to the target domain on the prediction result can be reduced, which can further improve the accuracy of the resource recommendation model. In addition, filtering out sample source domain interaction data that is not related to the target domain can also effectively control the amount of data that needs to be stored and processed, which is conducive to saving storage resources and processing resources.

[0377] Based on the above embodiment, the resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model. Figure 8 is a flowchart of another method for training a resource recommendation model provided in an embodiment of the present application. The embodiment of the present application is executed by a computer device, and the computer device is as follows Figure 1 The terminal 101 or the server 102 shown in FIG. Figure 8 , the method comprising:

[0378] 801. A computer device obtains sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object.

[0379] 802. The computer device obtains first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction matters, wherein the sample source domain interaction matters refer to the interaction between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction matters, wherein the sample target domain interaction matters refer to the interaction between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains.

[0380] 803. The computer device determines at least one sample source domain interaction item associated with the target domain in the first sample source domain interaction data, and the at least one sample source domain interaction item associated with the target domain constitutes the second sample source domain interaction data.

[0381] In one possible implementation, multiple target entities are determined, where a target entity refers to an entity to which resources in a target domain belong, and an entity refers to an object that provides resources; for any sample source domain interaction event in the first sample source domain interaction data, if the entity to which the resources in the sample source domain interaction event belong is any target entity, then it is determined that the sample source domain interaction event is associated with the target domain.

[0382] In another possible implementation, multiple target entities are determined, where the target entity refers to the entity to which the resources of the target domain belong, and the entity refers to the object that provides the resources; for any sample source domain interaction event in the first sample source domain interaction data, if the similarity between the entity to which the resources in the sample source domain interaction event belong and any target entity is greater than a similarity threshold, then it is determined that the sample source domain interaction event is associated with the target domain.

[0383] Optionally, target entity features of each target entity are obtained; features of entities to which resources in sample source domain interactive events belong are extracted to obtain sample source domain entity features; if the similarity between the sample source domain entity features and any target entity features is greater than a similarity threshold, it is determined that the sample source domain interactive event is associated with the target domain.

[0384] 804. The computer device extracts features from the sample object data through the feature extraction sub-model in the resource recommendation model to obtain sample object features.

[0385] 805. The computer device extracts features from the sample resource data through a feature extraction sub-model to obtain sample resource features.

[0386] 806. The computer device extracts features from the second sample source domain interaction data and the sample target domain interaction data through a feature extraction sub-model to obtain sample source domain interaction features and sample target domain interaction features.

[0387] In a possible implementation, the feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, and the first attention network includes a first mapping matrix. Step 806 includes the following steps 8061 to 8065.

[0388] 8061. Perform feature extraction on the second sample source domain interaction data through a feature extraction network to obtain initial sample source domain interaction features.

[0389] 8062. Through the feature extraction network, feature extraction is performed on the sample target domain interaction data to obtain initial sample target domain interaction features.

[0390] 8063. The initial sample source domain interaction feature and the initial sample target domain interaction feature are fused through a feature fusion network to obtain a first sample fused interaction feature.

[0391] 8064. Through the first attention network, the first sample fusion interaction feature is multiplied by the first mapping matrix to obtain the first sample query feature, the first sample key feature and the first sample value feature, and the first sample query feature, the first sample key feature and the first sample value feature are fused to obtain the second sample fusion interaction feature.

[0392] 8065. Based on the second sample fusion interaction feature, determine the sample source domain interaction feature and the sample target domain interaction feature.

[0393] Optionally, the feature extraction sub-model also includes a second attention network, and the second attention network includes a second mapping matrix.

[0394] The process of determining the sample source domain interaction feature in step 8065 includes: splitting the intermediate sample source domain interaction feature from the second sample fused interaction feature; multiplying the sample resource feature with the second mapping matrix through the second attention network to obtain the second sample query feature, and multiplying the intermediate sample source domain interaction feature with the second mapping matrix to obtain the second sample key feature and the second sample value feature; and fusing the second sample query feature, the second sample key feature and the second sample value feature through the second attention network to obtain the sample source domain interaction feature.

[0395] Optionally, the feature extraction sub-model also includes a third attention network, and the third attention network includes a third mapping matrix.

[0396] The process of determining the sample target domain interaction feature in step 8065 includes: splitting the intermediate sample target domain interaction feature from the second sample fused interaction feature; multiplying the sample resource feature with the third mapping matrix through the third attention network to obtain the third sample query feature, and multiplying the intermediate sample target domain interaction feature with the third mapping matrix to obtain the third sample key feature and the third sample value feature; and fusing the third sample query feature, the third sample key feature and the third sample value feature through the third attention network to obtain the sample target domain interaction feature.

[0397] 807. The computer device determines a recommendation prediction result based on sample object features, sample resource features, sample source domain interaction features, and sample target domain interaction features through a recommendation prediction sub-model in a resource recommendation model.

[0398] In an embodiment of the present application, the interaction between the sample object and the sample resource is predicted based on the characteristics of the sample object data, the characteristics of the sample resource data, the characteristics of the sample source domain interaction data, and the characteristics of the sample target domain interaction data. Through the potential characteristics of these sample data, the interest tendencies of the sample object and the degree of match between the interest tendencies of the sample object and the sample resource can be fully explored, thereby inferring the interaction tendency of the sample object with the sample resource, which is conducive to ensuring the accuracy of the recommendation prediction results and accelerating the training process of the resource recommendation model.

[0399] 808. The computer device trains the resource recommendation model based on the recommendation prediction results and the sample labels.

[0400] In one possible implementation, the computer device determines a first loss parameter based on the recommendation prediction result and the sample label, and trains the resource recommendation model based on the first loss parameter to reduce the first loss parameter obtained based on the trained resource recommendation model.

[0401] Optionally, the computer device uses the following formula to determine the first loss parameter.

[0402]

[0403] in, represents the first loss parameter, represents the recommendation prediction result, and y represents the sample label.

[0404] In one possible implementation, due to the cross-domain problem and the resource recommendation model is trained with the recommendation task in the target domain as the sample, the representation learning of the resources in the source domain is insufficient, and the gap between the resource features in the source domain and the resource features in the target domain is large. To overcome the above challenges, an additional supervised learning process can be added to the resource features in the source domain and the resource features in the target domain extracted by the resource recommendation model, so as to align the resource features in the source domain and the resource features in the target domain as the training target, and train the resource recommendation model. The following provides two training methods.

[0405] The first type: sample source domain interaction events include source domain resources and interaction time points between source domain resources, sample target domain interaction events include target domain resources and interaction time points between target domain resources, sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction event, and sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction event.

[0406] For any source domain resource in the second sample source domain interaction data, the computer device determines a first target domain resource in the sample target domain interaction data, and determines the source domain resource characteristics of the source domain resource and the target domain resource characteristics of the first target domain resource as a first positive sample pair, and the interaction time point of the first target domain resource is closest to the interaction time point of the source domain resource.

[0407] Then step 808 includes: training the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair.

[0408] In the interactive events of the same object in the same domain, the interactive resources should show similar characteristics. Furthermore, since the interests of the object in different fields will remain stable in the short term, it can be assumed that in the interactive events of the same object in different domains in a similar time period, the interactive resources should also show similar characteristics. Therefore, the source domain resource features and the target domain resource features with close interaction time points in the source domain and the target domain can be used as a positive sample pair, with the goal of shortening the distance between the two resource features in the positive sample pair.

[0409] Then, in step 808, the training goal is to reduce the difference between the recommendation prediction result and the sample label, and to improve the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair.

[0410] Optionally, the computer device determines, for any source domain resource in the second sample source domain interaction data, a second target domain resource in the sample target domain interaction data of any object other than the sample object, and determines the source domain resource characteristics of the source domain resource and the target domain resource characteristics of the second target domain resource as a first negative sample pair, and the second target domain resource is any target domain resource in the sample target domain interaction data of any object.

[0411] Then step 808 includes: training the resource recommendation model based on the recommendation prediction results, sample labels, the similarity between the source domain resource features and the target domain resource features in the first positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the first negative sample pair.

[0412] Correspondingly, in the interaction events of different objects in different domains, the interactive resources should show dissimilar characteristics. Therefore, the source domain resource characteristics of the object in the source domain and the target domain resource characteristics of other objects in the target domain can be used as a negative sample pair, with the goal of increasing the distance between the two resource characteristics in the negative sample pair as the training goal.

[0413] Exemplarily, the computer device determines a first loss parameter based on the recommendation prediction result and the sample label, determines a second loss parameter based on the similarity between the source domain resource features and the target domain resource features in the first positive sample pair and the similarity between the source domain resource features and the target domain resource features in the first negative sample pair, and trains the resource recommendation model based on the first loss parameter and the second loss parameter to reduce the first loss parameter and the second loss parameter obtained based on the trained resource recommendation model.

[0414] For example, the computer device determines the second loss parameter using the following formula.

[0415]

[0416] in, represents the second loss parameter, es uc represents the source domain resource features in the first positive sample pair, et uc represents the target domain resource feature in the first positive sample pair, es ic represents the source domain resource features in the first negative sample pair, et uo represents the target domain resource feature in the first negative sample pair, m represents the length of the source domain interaction feature, which can also be understood as the number of source domain resource features. τ is a hyperparameter, represents the source domain interaction feature. sim(·) represents the cosine similarity. uc represents the sample object. u represents all objects except the current training batch, and uo represents any object except the sample object.

[0417] In an embodiment of the present application, the resource recommendation model extracts source domain resource features of active domain resources and target domain resource features of target domain resources. Considering that resources of the same object interacting in different domains at a similar time will also be relatively close, the features of two resources with close interaction time points in the source domain and the target domain are used as positive sample pairs. The resource recommendation model is trained with the goal of narrowing the distance between the source domain resource features and the target domain resource features in the positive sample pairs. This can effectively align the resource features of different domains extracted by the resource recommendation model, improve the feature extraction capability of the resource recommendation model, and help ensure the accuracy of the cross-domain processing process.

[0418] In addition, considering that the resources interacted by the same object in different fields at different times will be quite different, the features of two resources in the source domain and the target domain with different interaction time points are used as negative sample pairs. The goal is to increase the distance between the source domain resource features and the target domain resource features in the negative sample pairs to train the resource recommendation model, which can further align the resource features of different domains extracted by the resource recommendation model.

[0419] The second type: sample source domain interaction events include source domain resources and entities to which the source domain resources belong, sample target domain interaction events include target domain resources and entities to which the target domain resources belong, sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction event, and sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction event.

[0420] For any source domain resource in the second sample source domain interaction data, the computer device determines a third target domain resource in the sample target domain interaction data, and determines the source domain resource characteristics of the source domain resource and the target domain resource characteristics of the third target domain resource as a second positive sample pair, and the entity to which the third target domain resource belongs is the same as the entity to which the source domain resource belongs.

[0421] Then step 808 includes: training the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the second positive sample pair.

[0422] Different resources with the same entity in the same domain should show similar features. It can be considered that this also applies to different resources across domains, that is, different resources with the same entity in different domains should also show similar features. Therefore, the source domain resource features and the target domain resource features belonging to the same entity in the source domain and the target domain can be used as a positive sample pair, with the goal of shortening the distance between the two resource features in the positive sample pair.

[0423] Then, in step 808, the training goal is to reduce the difference between the recommendation prediction result and the sample label, and to increase the similarity between the source domain resource features and the target domain resource features in the second positive sample pair.

[0424] Optionally, the computer device determines a fourth target domain resource for any source domain resource in the second sample source domain interaction data, and determines the source domain resource characteristics of the source domain resource and the target domain resource characteristics of the fourth target domain resource as a second negative sample pair, and the fourth target domain resource is any target domain resource that is different from the entity to which the source domain resource belongs.

[0425] Then step 808 includes: training the resource recommendation model based on the recommendation prediction results, sample labels, the similarity between the source domain resource features and the target domain resource features in the second positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the second negative sample pair.

[0426] Correspondingly, different resources with different entities in different domains should show dissimilar features. Therefore, the source domain resource features and target domain resource features of the object with different entities can be used as a negative sample pair, with the goal of increasing the distance between the two resource features in the negative sample pair as the training goal.

[0427] Exemplarily, the computer device determines a first loss parameter based on the recommendation prediction result and the sample label, determines a third loss parameter based on the similarity between the source domain resource features and the target domain resource features in the second positive sample pair and the similarity between the source domain resource features and the target domain resource features in the second negative sample pair, and trains the resource recommendation model based on the first loss parameter and the third loss parameter so that the first loss parameter and the third loss parameter obtained based on the trained resource recommendation model are reduced.

[0428] For example, the computer device determines the third loss parameter using the following formula.

[0429]

[0430] in, represents the third loss parameter, es lc represents the source domain resource features in the second positive sample pair, et lc represents the target domain resource features in the second positive sample pair, es lc represents the source domain resource features in the second negative sample pair, et lo represents the target domain resource feature in the second negative sample pair, m represents the length of the source domain interaction feature, which can also be understood as the number of source domain resource features. τ is a hyperparameter, represents the source domain interaction feature. sim(·) represents the cosine similarity.

[0431] In an embodiment of the present application, the resource recommendation model extracts source domain resource features of active domain resources and target domain resource features of target domain resources. Considering that resources belonging to the same entity in different domains are relatively close, the features of two resources belonging to the same entity in the source domain and the target domain are used as positive sample pairs. The resource recommendation model is trained with the goal of narrowing the distance between the source domain resource features and the target domain resource features in the positive sample pairs. This can effectively align the resource features of different domains extracted by the resource recommendation model, improve the feature extraction capability of the resource recommendation model, and help ensure the accuracy of the cross-domain processing process.

[0432] In addition, considering that the resources belonging to different entities in different fields may differ greatly, the features of two resources belonging to different entities in the source domain and the target domain are used as negative sample pairs. The resource recommendation model is trained with the goal of increasing the distance between the source domain resource features and the target domain resource features in the negative sample pairs. This can further align the resource features of different domains extracted by the resource recommendation model.

[0433] In one possible implementation, the computer device determines a target loss parameter based on the first loss parameter, the second loss parameter, and the third loss parameter, and trains a resource recommendation model based on the target loss parameter to reduce the target loss parameter obtained based on the trained resource recommendation model.

[0434] Optionally, the computer device uses the following formula to determine the target loss parameter.

[0435]

[0436]

[0437] in, represents the first loss parameter, represents the second loss parameter, represents the third loss parameter, represents the target loss parameter, λ uc , lc and λ ca It is a super parameter.

[0438] In a possible implementation, the above only takes the alignment of resource features of the source domain and the target domain as an example for explanation. In addition, the alignment of resource features of the same domain can also be used as a training target to train the resource recommendation model. The following provides the following multiple training methods.

[0439] Optionally, the computer device determines the first source domain resource and the second source domain resource in the second sample source domain interaction data of the sample object, and determines the source domain resource feature of the first source domain resource and the source domain resource feature of the second source domain resource as a third positive sample pair, the first source domain resource is any source domain resource, and the interaction time point of the second source domain resource is closest to the interaction time point of the first source domain resource. The computer device trains the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the two source domain resource features in the third positive sample pair. The training goal is to reduce the difference between the recommendation prediction result and the sample label, and to increase the similarity between the two source domain resource features in the third positive sample pair.

[0440] Optionally, for the first source domain resource in the second sample source domain interaction data, the computer device determines a third source domain resource in the sample source domain interaction data of any object other than the sample object, and determines the source domain resource features of the first source domain resource and the source domain resource features of the third source domain resource as a third negative sample pair, where the third source domain resource is any source domain resource in the sample source domain interaction data of any object. The computer device trains the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the two source domain resource features in the third negative sample pair. The training goal is to reduce the difference between the recommendation prediction result and the sample label, and to reduce the similarity between the two source domain resource features in the third negative sample pair.

[0441] Optionally, the computer device determines the first source domain resource and the fourth source domain resource in the second sample source domain interaction data of the sample object, and determines the source domain resource features of the first source domain resource and the source domain resource features of the fourth source domain resource as a fourth positive sample pair, the first source domain resource is any source domain resource, and the entity to which the fourth source domain resource belongs is the same as the entity to which the first source domain resource belongs. The computer device trains the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the two source domain resource features in the fourth positive sample pair. The training goal is to reduce the difference between the recommendation prediction result and the sample label, and to increase the similarity between the two source domain resource features in the fourth positive sample pair.

[0442] Optionally, the computer device determines the fifth source domain resource for the first source domain resource in the second sample source domain interaction data, and determines the source domain resource features of the first source domain resource and the source domain resource features of the fifth source domain resource as a fourth negative sample pair, where the fifth source domain resource is any source domain resource that is different from the entity to which the first source domain resource belongs. The computer device trains the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the two source domain resource features in the fourth negative sample pair. The training goal is to reduce the difference between the recommendation prediction result and the sample label, and to reduce the similarity between the two source domain resource features in the fourth negative sample pair.

[0443] The above only takes the process of aligning the resource features of the source domain as an example. The process of aligning the resource features of the target domain is the same as that of aligning the resource features of the source domain, which will not be repeated here.

[0444] The method provided in the embodiment of the present application introduces sample source domain interaction data, so it can make up for the defect of insufficient sample target domain interaction data, alleviate the impact caused by the sparse sample target domain interaction data, and help improve the accuracy of the resource recommendation model. On the other hand, since only the sample source domain interaction data associated with the target domain is considered, the sample source domain interaction data irrelevant to the target domain is discarded, so the interference caused by the sample source domain interaction data irrelevant to the target domain on the prediction results can be reduced, which can further improve the accuracy of the resource recommendation model. In addition, filtering out the sample source domain interaction data irrelevant to the target domain can also effectively control the amount of data required to be stored and processed, which is conducive to saving storage resources and processing resources.

[0445] Fig. 9 is a schematic diagram of a training method for a resource recommendation model provided in an embodiment of the present application, such as Fig. 9 As shown, the initial samples are processed from the user logs, and the initial samples are processed to obtain training samples. The source domain interaction data is obtained from the source domain, the target domain interaction data is obtained from the target domain, the source domain interaction data is cross-domain screened, and the source domain interaction data associated with the target domain is obtained. The screened source domain interaction data and target domain interaction data are feature extracted and stored in the feature library. The resource recommendation model is trained using the training samples and the data in the feature library. Among them, the data stored in the feature library can also be used to recommend online resources.

[0446] Fig.10 is an overall architecture diagram of a resource recommendation method provided in an embodiment of the present application, such as Fig.10 As shown in the figure, the overall architecture includes a cross-domain retrieval module, a cross-domain feature alignment module, a feature enhancement module and a recommendation prediction module.

[0447] Among them, the cross-domain retrieval module is used to filter the source domain interaction data associated with the target domain. For example, the target entity includes entity A, entity B and entity F. Through cross-domain retrieval, it can be determined that the source domain interaction items corresponding to entity A, entity B and entity D in the source domain are associated with the target domain.

[0448] Among them, the cross-domain feature alignment module is used to align the resource features of the source domain and the target domain. For the same object, the source domain resource features and the target domain resource features with similar interaction time points in the source domain and the target domain need to be brought closer. For the same entity, the source domain resource features and the target domain resource features that belong to the same entity in the source domain and the target domain need to be brought closer.

[0449] Among them, the feature enhancement module is used to enhance the source domain interaction features associated with the target domain, and to enhance the source domain interaction features associated with the target resource. The attention mechanism is used to process the source domain interaction features and the target domain interaction features, the attention mechanism is used to process the source domain interaction features and the resource features, and the attention mechanism is used to process the target domain interaction features and the resource features.

[0450] The recommendation prediction module is used to determine the recommendation prediction result according to the object features, resource features, source domain interaction features and target domain interaction features. For example, the recommendation prediction module includes a recommendation prediction sub-model, and the recommendation prediction sub-model includes a splicing network and a prediction network. The object features, resource features, source domain interaction features and target domain interaction features are input into the splicing network for splicing to obtain the spliced ​​features, and the spliced ​​features are input into the prediction network to obtain the recommendation prediction result.

[0451] In order to verify the effectiveness of the resource recommendation method provided in the embodiment of the present application, experimental data were used to test the resource recommendation method in the embodiment of the present application and the resource recommendation method provided by the related technology. The test results are shown in Table 1 below.

[0452] Table 1

[0453] Model Name AUC CTR Related technologies 1 0.8304 base Related Technology 2 0.8371 +2.04% This application 0.8463 +4.69%

[0454] The offline evaluation index uses AUG (Area Under Curve, a model performance measurement index), and the online evaluation index uses CTR. As can be seen from Table 1, the solution of the embodiment of the present application achieves better results than related technologies both offline and online.

[0455] The embodiment of the present application proposes a cross-domain ultra-long interaction data modeling method based on retrieval and alignment, which is used to solve the resource recommendation task of applying rich interaction data in the source domain to the target domain, so as to improve the accuracy and diversity of resource recommendation. The method of the embodiment of the present application can be applied to any scenario where resource recommendation is required.

[0456] For example, the source domain is the shopping domain, the target domain is the advertising domain, the resources in the source domain are items in the shopping domain, and the resources in the target domain are virtual resources placed in the advertising domain, and the virtual resources are used to exchange items in the source domain. For example, the virtual resources are coupons or exchange vouchers.

[0457] Optionally, the virtual resources in the target domain and the items in the source domain may be provided by the same entity or by different entities. Optionally, the virtual resources provided by each entity in the target domain are used to exchange the items provided by each entity in the source domain. For example, the virtual resources provided by entity A in the target domain are used to exchange the items provided by entity A in the source domain, but the virtual resources provided by entity A in the target domain cannot be used to exchange the items provided by other entities in the source domain except entity A, and the items provided by entity A in the source domain cannot be exchanged using the virtual resources provided by other entities in the target domain except entity A.

[0458] The following uses the example of the source domain being the shopping domain and the target domain being the advertising domain to illustrate the process of the resource recommendation method provided in the embodiment of the present application. Fig.11 is a flowchart of another resource recommendation method provided in an embodiment of the present application, see Fig.11 , the method comprising:

[0459] 1101. When an object meets an advertisement delivery condition, the terminal sends a resource recommendation task for the object to the server.

[0460] 1102. In response to the resource recommendation task for the object, the server obtains object data of the object and virtual resource data of a target virtual resource, where the target virtual resource is used to exchange items.

[0461] 1103. The server obtains the first item interaction data of the object and the virtual resource interaction data of the object.

[0462] The first item interaction data includes a plurality of item interaction events, where an item interaction event refers to an interaction between an object and an item in the shopping field, for example, an interaction behavior between an object and an item refers to an behavior of purchasing an item.

[0463] The virtual resource interaction data includes multiple virtual resource interaction events, and the virtual resource interaction event refers to the interaction between the object and the virtual resource placed in the advertising placement field. For example, the virtual resource is a coupon placed in the advertising placement field, and the interaction behavior between the object and the virtual resource refers to the behavior of receiving the coupon.

[0464] 1104. The server determines at least one item interaction event associated with the advertising delivery field in the first item interaction data, and the at least one item interaction event associated with the advertising delivery field constitutes the second item interaction data.

[0465] 1105. The server determines a recommendation prediction result of the target virtual resource based on the object data, the virtual resource data, the second item interaction data and the virtual resource interaction data. The recommendation prediction result is used to characterize the interaction between the object and the target virtual resource if the target virtual resource is recommended to the object.

[0466] 1106. If the recommendation prediction result meets the recommendation condition, the server returns the target virtual resource to the terminal.

[0467] 1107. The terminal recommends the target virtual resource to the object.

[0468] In a possible implementation, the terminal runs a target client, which provides multiple services, such as payment services and advertising services, etc. For example, after the user completes the payment operation on the target client, advertising resources are recommended to the user, and the advertising resources include virtual resources for exchanging items.

[0469] Fig.12 is a schematic diagram of a service interface provided by an embodiment of the present application, such as Fig.12 As shown, the service interface provides a recharge entrance 1201. After the user clicks the recharge entrance 1201, he / she will be redirected to the recharge interface, which is used to recharge the phone bill. In addition, the recharge interface also includes a benefit collection entrance 1202.

[0470] Optionally, after completing the phone charge recharge, Fig.12 The recharge interface shown jumps to Fig.13 The recharge interface shown is as follows: Fig.13 As shown, the recharge interface displays a message notification of successful recharge. In addition, the recharge interface also includes advertising resources 1301 recommended to the user. The advertising resources 1301 carry virtual resources for exchanging items, such as virtual resources such as coupons.

[0471] Optionally, if the user Fig.12 Click on the benefit collection entrance 1202 on the top-up interface shown, and you will be redirected to Fig.14 The service interface shown, Fig.14 The service interface shown displays multiple advertising resources recommended for users, each of which carries virtual resources for exchanging items, such as coupons, red envelopes, vouchers, discount coupons, etc.

[0472] The above description is only based on the shopping and advertising fields as examples. In addition, the method of the embodiment of the present application can also be applied to other resource recommendation scenarios. For example, e-books are used as the source domain and movies are used as the target domain. Or, self-selected stocks are used as the source domain and stock information is used as the target domain.

[0473] Fig.15 is a schematic diagram of the structure of a resource recommendation device provided in an embodiment of the present application. Fig.15 , the device comprises:

[0474] A first acquisition module 1501 is used to acquire object data of an object and resource data of a target resource, where the target resource belongs to a target domain;

[0475] The second acquisition module 1502 is provided with first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction items, wherein the source domain interaction items refer to the interaction between the object and the resources in the source domain, and the target domain interaction data includes a plurality of target domain interaction items, wherein the target domain interaction items refer to the interaction between the object and the resources in the target domain, and the source domain and the target domain belong to different fields;

[0476] A screening module 1503 is used to determine at least one source domain interaction item associated with the target domain in the first source domain interaction data, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data;

[0477] The prediction module 1504 is used to determine the recommendation prediction result of the target resource based on the object data, resource data, the second source domain interaction data and the target domain interaction data. The recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

[0478] The resource recommendation device provided in the embodiment of the present application needs to refer to the historical interaction between the object and other resources to infer the object's interaction tendency with the target resource when determining whether to recommend the target resource of the target domain to the object. Therefore, the target domain interaction data of the object in the target domain is obtained. Considering that the target domain interaction data may have insufficient data, the source domain interaction data of the object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the source domain interaction data. Then, the target domain interaction data and the screened source domain interaction data are used as references to predict the interaction between the object and the target resource if the target resource is recommended to the object. On the one hand, since the source domain interaction data is introduced, the defect of insufficient target domain interaction data can be compensated, and the impact caused by the sparse target domain interaction data is alleviated. On the other hand, since only the source domain interaction data associated with the target domain is considered, the source domain interaction data irrelevant to the target domain is discarded, so the interference of the source domain interaction data irrelevant to the target domain on the prediction result can be reduced, thereby improving the accuracy of resource recommendation as a whole.

[0479] Alternatively, see Fig.16 , screening module 1503, for:

[0480] Determine multiple target entities, where the target entity refers to the entity to which the resources of the target domain belong, and the entity refers to the object that provides the resources;

[0481] For any source domain interaction event in the first source domain interaction data, if the entity to which the resource in the source domain interaction event belongs is any target entity, it is determined that the source domain interaction event is associated with the target domain.

[0482] Alternatively, see Fig.16 , screening module 1503, for:

[0483] Determine multiple target entities, where the target entity refers to the entity to which the resources of the target domain belong, and the entity refers to the object that provides the resources;

[0484] For any source domain interaction event in the first source domain interaction data, if the similarity between an entity to which a resource in the source domain interaction event belongs and any target entity is greater than a similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0485] Alternatively, see Fig.16 , screening module 1503, for:

[0486] Obtain target entity features for each target entity;

[0487] Extract features of entities to which resources belong in source domain interactive events to obtain source domain entity features;

[0488] If the similarity between the source domain entity feature and any target entity feature is greater than the similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

[0489] Alternatively, see Fig.16 , prediction module 1504, for:

[0490] Extract features from object data to obtain object features;

[0491] Extract features from resource data to obtain resource features;

[0492] Extracting features from the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features;

[0493] Based on object features, resource features, source domain interaction features and target domain interaction features, the recommendation prediction results of the target resources are determined.

[0494] Alternatively, see Fig.16 , prediction module 1504, for:

[0495] Extracting features from the second source domain interaction data to obtain initial source domain interaction features, and extracting features from the target domain interaction data to obtain initial target domain interaction features;

[0496] The initial source domain interaction feature and the initial target domain interaction feature are fused to obtain a first fused interaction feature;

[0497] Multiplying the first fused interaction feature by the first mapping matrix to obtain a first query feature, a first key feature, and a first value feature;

[0498] Fusing the first query feature, the first key feature, and the first value feature to obtain a second fused interactive feature;

[0499] Based on the second fused interactive features, source domain interactive features and target domain interactive features are determined.

[0500] Alternatively, see Fig.16 , prediction module 1504, for:

[0501] Separating the intermediate source domain interaction features from the second fused interaction features;

[0502] Multiplying the resource feature by the second mapping matrix to obtain a second query feature;

[0503] Multiplying the intermediate source domain interaction feature with the second mapping matrix to obtain a second key feature and a second value feature;

[0504] The second query feature, the second key feature and the second value feature are fused to obtain the source domain interaction feature.

[0505] Alternatively, see Fig.16 , prediction module 1504, for:

[0506] Separating the intermediate target domain interaction features from the second fused interaction features;

[0507] Multiplying the resource feature by the third mapping matrix to obtain a third query feature;

[0508] Multiply the intermediate target domain interaction feature by the third mapping matrix to obtain a third key feature and a third value feature;

[0509] The third query feature, the third key feature and the third value feature are fused to obtain the target domain interaction feature.

[0510] Alternatively, see Fig.16 The resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model. The prediction module 1504 is used to:

[0511] Through the feature extraction sub-model, feature extraction is performed on the object data to obtain object features;

[0512] Through the feature extraction sub-model, feature extraction is performed on resource data to obtain resource features;

[0513] Through the feature extraction sub-model, feature extraction is performed on the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features;

[0514] Through the recommendation prediction sub-model, the recommendation prediction results of the target resources are determined based on object features, resource features, source domain interaction features and target domain interaction features.

[0515] Alternatively, see Fig.16 The feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, and the first attention network includes a first mapping matrix; a prediction module 1504 is used to:

[0516] Through the feature extraction network, feature extraction is performed on the second source domain interaction data to obtain the initial source domain interaction features, and feature extraction is performed on the target domain interaction data to obtain the initial target domain interaction features;

[0517] The initial source domain interaction feature and the initial target domain interaction feature are fused through a feature fusion network to obtain a first fused interaction feature;

[0518] Through the first attention network, the first fused interactive feature is multiplied by the first mapping matrix to obtain a first query feature, a first key feature and a first value feature, and the first query feature, the first key feature and the first value feature are fused to obtain a second fused interactive feature;

[0519] Based on the second fused interactive features, source domain interactive features and target domain interactive features are determined.

[0520] Alternatively, see Fig.16 , the feature extraction sub-model further includes a second attention network, the second attention network includes a second mapping matrix; a prediction module 1504, for:

[0521] Separating the intermediate source domain interaction features from the second fused interaction features;

[0522] Through the second attention network, the resource feature is multiplied by the second mapping matrix to obtain the second query feature, and the intermediate source domain interaction feature is multiplied by the second mapping matrix to obtain the second key feature and the second value feature;

[0523] Through the second attention network, the second query feature, the second key feature and the second value feature are fused to obtain the source domain interaction feature.

[0524] Alternatively, see Fig.16 , the feature extraction sub-model further includes a third attention network, the third attention network includes a third mapping matrix; a prediction module 1504, for:

[0525] Separating the intermediate target domain interaction features from the second fused interaction features;

[0526] Through the third attention network, the resource feature is multiplied by the third mapping matrix to obtain the third query feature, and the intermediate target domain interaction feature is multiplied by the third mapping matrix to obtain the third key feature and the third value feature;

[0527] Through the third attention network, the third query feature, the third key feature and the third value feature are fused to obtain the target domain interaction feature.

[0528] Alternatively, see Fig.16 , the device further comprises:

[0529] The recommendation module 1505 is used to recommend the target resource to the object if the recommendation prediction result meets the recommendation condition.

[0530] It should be noted that the resource recommendation device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the resource recommendation device provided in the above embodiment and the resource recommendation method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0531] Fig.17 Schematic diagram of a training device for a resource recommendation model provided in an embodiment of the present application. Fig.17 , the device comprises:

[0532] The first acquisition module 1701 is used to acquire sample object data of the sample object, sample resource data of the sample resource, and a sample label, where the sample resource belongs to the target domain, and the sample label is used to indicate the actual interaction between the sample object and the sample resource after the sample resource is recommended to the sample object;

[0533] The second acquisition module 1702 is used to acquire first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to the interaction between the sample object and the resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to the interaction between the sample object and the resources in the target domain, and the source domain and the target domain belong to different fields;

[0534] A screening module 1703 is used to determine at least one sample source domain interaction item associated with the target domain in the first sample source domain interaction data, and the at least one sample source domain interaction item associated with the target domain constitutes the second sample source domain interaction data;

[0535] The prediction module 1704 is used to determine the recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data and the target domain sample interaction data through the resource recommendation model, and the recommendation prediction result is used to represent the predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object;

[0536] The training module 1705 is used to train the resource recommendation model based on the recommendation prediction results and sample labels.

[0537] The training device of the resource recommendation model provided in the embodiment of the present application, when training the resource recommendation model, for the recommendation task of the target domain, obtains the sample target domain interaction data of the sample object in the target domain, considering that the sample target domain interaction data may have insufficient data, so the sample source domain interaction data of the sample object in the source domain is additionally obtained, and the part associated with the target domain is screened out from the sample source domain interaction data, and then the sample target domain interaction data and the screened sample source domain interaction data are used as references, and the resource recommendation model is used to predict the recommendation prediction result of recommending the sample resource to the sample object, and then the resource recommendation model is trained using the recommendation prediction result and the sample label. On the one hand, due to the introduction of the sample source domain interaction data, the defect of insufficient sample target domain interaction data can be compensated, the influence caused by the sparse sample target domain interaction data is alleviated, and it is conducive to improving the accuracy of the resource recommendation model. On the other hand, since only the sample source domain interaction data associated with the target domain is considered, the sample source domain interaction data irrelevant to the target domain is discarded, so the interference caused by the sample source domain interaction data irrelevant to the target domain on the prediction result can be reduced, which can further improve the accuracy of the resource recommendation model. In addition, filtering out sample source domain interaction data that is not related to the target domain can also effectively control the amount of data that needs to be stored and processed, which is conducive to saving storage resources and processing resources.

[0538] Alternatively, see Fig.18 The resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model; the prediction module 1704 is used to:

[0539] Through the feature extraction sub-model, feature extraction is performed on the sample object data to obtain the sample object features;

[0540] Through the feature extraction sub-model, feature extraction is performed on the sample resource data to obtain sample resource features;

[0541] Through the feature extraction sub-model, feature extraction is performed on the second sample source domain interaction data and the sample target domain interaction data to obtain the sample source domain interaction feature and the sample target domain interaction feature;

[0542] The recommendation prediction sub-model is used to determine the recommendation prediction result based on sample object characteristics, sample resource characteristics, sample source domain interaction characteristics, and sample target domain interaction characteristics.

[0543] Alternatively, see Fig.18The sample source domain interaction items include the interaction time points of source domain resources and source domain resources, the sample target domain interaction items include the interaction time points of target domain resources and target domain resources, the sample source domain interaction features include the source domain resource features of the source domain resources in each sample source domain interaction item, and the sample target domain interaction features include the target domain resource features of the target domain resources in each sample target domain interaction item; the device also includes:

[0544] A first determination module 1706 is configured to determine, for any source domain resource in the second sample source domain interaction data, a first target domain resource in the sample target domain interaction data, and determine the source domain resource feature of the source domain resource and the target domain resource feature of the first target domain resource as a first positive sample pair, wherein the interaction time point of the first target domain resource is closest to the interaction time point of the source domain resource;

[0545] The training module 1705 is used to:

[0546] The resource recommendation model is trained based on the recommendation prediction results, sample labels, and the similarity between the source domain resource features and the target domain resource features in the first positive sample pair.

[0547] Alternatively, see Fig.18 , the device further comprises:

[0548] The second determination module 1707 is used to determine, for any source domain resource in the second sample source domain interaction data, a second target domain resource in the sample target domain interaction data of any object other than the sample object, and determine the source domain resource feature of the source domain resource and the target domain resource feature of the second target domain resource as a first negative sample pair, and the second target domain resource is any target domain resource in the sample target domain interaction data of any object;

[0549] The training module 1705 is used to:

[0550] The resource recommendation model is trained based on the recommendation prediction results, sample labels, the similarity between the source domain resource features and the target domain resource features in the first positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the first negative sample pair.

[0551] Alternatively, see Fig.18 The sample source domain interaction items include source domain resources and entities to which the source domain resources belong, the sample target domain interaction items include target domain resources and entities to which the target domain resources belong, the sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction item, and the sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction item; the device also includes:

[0552] A third determination module 1708 is configured to determine, for any source domain resource in the second sample source domain interaction data, a third target domain resource in the sample target domain interaction data, and determine the source domain resource feature of the source domain resource and the target domain resource feature of the third target domain resource as a second positive sample pair, wherein the entity to which the third target domain resource belongs is the same as the entity to which the source domain resource belongs;

[0553] The training module 1705 is used to:

[0554] The resource recommendation model is trained based on the recommendation prediction results, sample labels, and the similarity between the source domain resource features and the target domain resource features in the second positive sample pair.

[0555] Alternatively, see Fig.18 , the device further comprises:

[0556] The fourth determination module 1709 is used to determine a fourth target domain resource for any source domain resource in the second sample source domain interaction data, and determine the source domain resource feature of the source domain resource and the target domain resource feature of the fourth target domain resource as a second negative sample pair, wherein the fourth target domain resource is any target domain resource different from the entity to which the source domain resource belongs;

[0557] The training module 1705 is used to:

[0558] The resource recommendation model is trained based on the recommendation prediction results, sample labels, the similarity between the source domain resource features and the target domain resource features in the second positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the second negative sample pair.

[0559] Alternatively, see Fig.18 The feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, and the first attention network includes a first mapping matrix; the prediction module 1704 is used to:

[0560] Through the feature extraction network, feature extraction is performed on the second sample source domain interaction data to obtain the initial sample source domain interaction feature, and feature extraction is performed on the sample target domain interaction data to obtain the initial sample target domain interaction feature;

[0561] The initial sample source domain interaction feature and the initial sample target domain interaction feature are fused through a feature fusion network to obtain a first sample fused interaction feature;

[0562] Through the first attention network, the first sample fusion interaction feature is multiplied by the first mapping matrix to obtain the first sample query feature, the first sample key feature and the first sample value feature, and the first sample query feature, the first sample key feature and the first sample value feature are fused to obtain the second sample fusion interaction feature;

[0563] Based on the second sample fusion interaction feature, the sample source domain interaction feature and the sample target domain interaction feature are determined.

[0564] Alternatively, see Fig.18 , the feature extraction sub-model further includes a second attention network, the second attention network includes a second mapping matrix; a prediction module 1704, for:

[0565] Separate the source domain interaction features of the intermediate samples from the fusion interaction features of the second samples;

[0566] Through the second attention network, the sample resource feature is multiplied by the second mapping matrix to obtain the second sample query feature, and the intermediate sample source domain interaction feature is multiplied by the second mapping matrix to obtain the second sample key feature and the second sample value feature;

[0567] Through the second attention network, the second sample query feature, the second sample key feature and the second sample value feature are fused to obtain the sample source domain interaction feature.

[0568] Alternatively, see Fig.18 , the feature extraction sub-model further includes a third attention network, the third attention network includes a third mapping matrix; a prediction module 1704, for:

[0569] Separating the target domain interaction features of the intermediate samples from the fused interaction features of the second samples;

[0570] Through the third attention network, the sample resource feature is multiplied by the third mapping matrix to obtain the third sample query feature, and the intermediate sample target domain interaction feature is multiplied by the third mapping matrix to obtain the third sample key feature and the third sample value feature;

[0571] Through the third attention network, the third sample query feature, the third sample key feature and the third sample value feature are fused to obtain the sample target domain interaction feature.

[0572] Alternatively, see Fig.18 , screening module 1703, for:

[0573] Determine multiple target entities, where the target entity refers to the entity to which the resources of the target domain belong, and the entity refers to the object that provides the resources;

[0574] For any sample source domain interaction event in the first sample source domain interaction data, if the entity to which the resource in the sample source domain interaction event belongs is any target entity, it is determined that the sample source domain interaction event is associated with the target domain.

[0575] Alternatively, see Fig.18 , screening module 1703, for:

[0576] Determine multiple target entities, where the target entity refers to the entity to which the resources of the target domain belong, and the entity refers to the object that provides the resources;

[0577] For any sample source domain interaction event in the first sample source domain interaction data, if the similarity between the entity to which the resource in the sample source domain interaction event belongs and any target entity is greater than a similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0578] Alternatively, see Fig.18 , screening module 1703, for:

[0579] Obtain target entity features for each target entity;

[0580] Extract features of entities to which resources belong in interactive events in the sample source domain to obtain entity features in the sample source domain;

[0581] If the similarity between the sample source domain entity feature and any target entity feature is greater than the similarity threshold, it is determined that the sample source domain interaction event is associated with the target domain.

[0582] It should be noted that the resource recommendation device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the training device for the resource recommendation model provided in the above embodiment and the training method embodiment of the resource recommendation model belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0583] An embodiment of the present application also provides a computer device, which includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed in the resource recommendation method of the above embodiment.

[0584] Optionally, the computer device is provided as a terminal. Fig.19 A schematic diagram of the structure of a terminal 1900 provided by an exemplary embodiment of the present application is shown.

[0585] The terminal 1900 includes a processor 1901 and a memory 1902 .

[0586] The processor 1901 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1901 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1901 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 1901 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1901 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0587] The memory 1902 may include one or more computer-readable storage media, which may be non-transitory. The memory 1902 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1902 is used to store at least one computer program, which is used by the processor 1901 to implement the resource recommendation method or resource recommendation model training method provided in the method embodiment of the present application.

[0588] In some embodiments, the terminal 1900 may further optionally include: a peripheral device interface 1903 and at least one peripheral device. The processor 1901, the memory 1902 and the peripheral device interface 1903 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 1903 via a bus, a signal line or a circuit board. Optionally, the peripheral device includes: at least one of a radio frequency circuit 1904, a display screen 1905, a camera assembly 1906, an audio circuit 1907 and a power supply 1908.

[0589] The peripheral device interface 1903 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1901 and the memory 1902. In some embodiments, the processor 1901, the memory 1902, and the peripheral device interface 1903 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1901, the memory 1902, and the peripheral device interface 1903 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0590] The radio frequency circuit 1904 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1904 communicates with a communication network and other communication devices via electromagnetic signals. The radio frequency circuit 1904 converts an electrical signal into an electromagnetic signal for transmission, or converts a received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1904 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency circuit 1904 can communicate with other devices via at least one wireless communication protocol.

[0591] The display screen 1905 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1905 is a touch display screen, the display screen 1905 also has the ability to collect touch signals on the surface of the display screen 1905 or above the surface. The touch signal can be input to the processor 1901 as a control signal for processing. At this time, the display screen 1905 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1905 can be one, set on the front panel of the terminal 1900; in other embodiments, the display screen 1905 can be at least two, respectively set on different surfaces of the terminal 1900 or in a folding design; in other embodiments, the display screen 1905 can be a flexible display screen, set on the curved surface or folding surface of the terminal 1900.

[0592] The camera assembly 1906 is used to capture images or videos. Optionally, the camera assembly 1906 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal 1900, and the rear camera is arranged on the back of the terminal 1900. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1906 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0593] The audio circuit 1907 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 1901 for processing, or input them into the radio frequency circuit 1904 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1900. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 1901 or the radio frequency circuit 1904 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 1907 may also include a headphone jack.

[0594] The power supply 1908 is used to power various components in the terminal 1900. The power supply 1908 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 1908 includes a rechargeable battery, the rechargeable battery can support wired charging or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0595] Those skilled in the art will understand that Fig.19 The structure shown in the figure does not constitute a limitation on the terminal 1900, and the terminal 1900 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0596] Optionally, the computer device is provided as a server. Fig. 20This is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 2000 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 2001 and one or more memories 2002, wherein the memory 2002 stores at least one computer program, and the at least one computer program is loaded and executed by the processor 2001 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for implementing device functions, which will not be described in detail here.

[0597] An embodiment of the present application also provides a computer-readable storage medium, which stores at least one computer program, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the resource recommendation method of the above embodiment, or to implement the operations performed by the training method of the resource recommendation model of the above embodiment.

[0598] An embodiment of the present application also provides a computer program product, including a computer program, which is loaded and executed by a processor to implement the operations performed by the resource recommendation method of the above embodiment, or to implement the operations performed by the training method of the resource recommendation model of the above embodiment.

[0599] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0600] The above description is only an optional embodiment of the embodiments of the present application and is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A resource recommendation method, characterized in that: The method comprises: Obtaining object data of an object and resource data of a target resource, the target resource belonging to a target domain; Acquire first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction items, wherein the source domain interaction items refer to the interaction between the object and resources in the source domain, and the target domain interaction data includes a plurality of target domain interaction items, wherein the target domain interaction items refer to the interaction between the object and resources in the target domain, and the source domain and the target domain belong to different fields; In the first source domain interaction data, at least one source domain interaction item associated with the target domain is determined, and the at least one source domain interaction item associated with the target domain constitutes the second source domain interaction data; Based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, a recommendation prediction result of the target resource is determined, and the recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

2. The method according to claim 1, characterized in that Determining at least one source domain interaction item associated with the target domain in the first source domain interaction data includes: Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources; For any source domain interaction event in the first source domain interaction data, if the entity to which the resource in the source domain interaction event belongs is any of the target entities, it is determined that the source domain interaction event is associated with the target domain.

3. The method according to claim 1, characterized in that Determining at least one source domain interaction item associated with the target domain in the first source domain interaction data includes: Determine a plurality of target entities, wherein the target entity refers to an entity to which the resources of the target domain belong, and the entity refers to an object that provides resources; For any source domain interaction event in the first source domain interaction data, if the similarity between the entity to which the resource in the source domain interaction event belongs and any of the target entities is greater than a similarity threshold, it is determined that the source domain interaction event is associated with the target domain.

4. The method according to claim 1, characterized in that: The determining the recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data includes: Extracting features from the object data to obtain object features; Extracting features from the resource data to obtain resource features; Performing feature extraction on the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features; Based on the object features, the resource features, the source domain interaction features, and the target domain interaction features, a recommendation prediction result of the target resource is determined.

5. The method according to claim 4, characterized in that The extracting features of the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features includes: Performing feature extraction on the second source domain interaction data to obtain initial source domain interaction features, and performing feature extraction on the target domain interaction data to obtain initial target domain interaction features; Fusing the initial source domain interaction feature and the initial target domain interaction feature to obtain a first fused interaction feature; Multiplying the first fused interaction feature by the first mapping matrix to obtain a first query feature, a first key feature, and a first value feature; Fusing the first query feature, the first key feature, and the first value feature to obtain a second fused interaction feature; Based on the second fused interaction feature, the source domain interaction feature and the target domain interaction feature are determined.

6. The method according to claim 5, characterized in that The determining the source domain interaction feature and the target domain interaction feature based on the second fused interaction feature includes: Separating the intermediate source domain interaction features from the second fused interaction features; Multiplying the resource feature by a second mapping matrix to obtain a second query feature; Multiplying the intermediate source domain interaction feature by the second mapping matrix to obtain a second key feature and a second value feature; The second query feature, the second key feature and the second value feature are fused to obtain the source domain interaction feature.

7. The method according to claim 5, characterized in that The determining the source domain interaction feature and the target domain interaction feature based on the second fused interaction feature includes: Separating an intermediate target domain interaction feature from the second fused interaction feature; Multiplying the resource feature by a third mapping matrix to obtain a third query feature; Multiplying the intermediate target domain interaction feature with the third mapping matrix to obtain a third key feature and a third value feature; The third query feature, the third key feature and the third value feature are fused to obtain the target domain interaction feature.

8. The method according to any one of claims 1 to 7, characterized in that: The resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model. The method of determining the recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data includes: Performing feature extraction on the object data through the feature extraction sub-model to obtain object features; By using the feature extraction sub-model, feature extraction is performed on the resource data to obtain resource features; By using the feature extraction sub-model, feature extraction is performed on the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features; The recommendation prediction sub-model is used to determine the recommendation prediction result of the target resource based on the object feature, the resource feature, the source domain interaction feature and the target domain interaction feature.

9. The method according to claim 8, characterized in that The feature extraction sub-model includes a feature extraction network, a feature fusion network and a first attention network, wherein the first attention network includes a first mapping matrix; the feature extraction sub-model is used to extract features from the second source domain interaction data and the target domain interaction data to obtain source domain interaction features and target domain interaction features, including: Through the feature extraction network, feature extraction is performed on the second source domain interaction data to obtain initial source domain interaction features, and feature extraction is performed on the target domain interaction data to obtain initial target domain interaction features; The initial source domain interaction feature and the initial target domain interaction feature are fused through the feature fusion network to obtain a first fused interaction feature; multiplying the first fused interactive feature by the first mapping matrix through the first attention network to obtain a first query feature, a first key feature, and a first value feature, and fusing the first query feature, the first key feature, and the first value feature to obtain a second fused interactive feature; Based on the second fused interaction feature, the source domain interaction feature and the target domain interaction feature are determined.

10. A method for training a resource recommendation model, characterized in that: The method comprises: Acquire sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object; Acquire first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to interactions between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to interactions between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains; In the first sample source domain interaction data, at least one sample source domain interaction event associated with the target domain is determined, and the at least one sample source domain interaction event associated with the target domain constitutes the second sample source domain interaction data; Determine, by means of a resource recommendation model, a recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data, and the target domain sample interaction data, wherein the recommendation prediction result is used to characterize the predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object; The resource recommendation model is trained based on the recommendation prediction result and the sample label.

11. The method according to claim 10, characterized in that The resource recommendation model includes a feature extraction sub-model and a recommendation prediction sub-model; the resource recommendation model is used to determine the recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data and the target domain sample interaction data, including: Performing feature extraction on the sample object data through the feature extraction sub-model to obtain sample object features; By using the feature extraction sub-model, feature extraction is performed on the sample resource data to obtain sample resource features; By using the feature extraction sub-model, feature extraction is performed on the second sample source domain interaction data and the sample target domain interaction data to obtain sample source domain interaction features and sample target domain interaction features; The recommendation prediction result is determined through the recommendation prediction sub-model based on the sample object characteristics, the sample resource characteristics, the sample source domain interaction characteristics and the sample target domain interaction characteristics.

12. The method according to claim 11, characterized in that The sample source domain interaction items include source domain resources and interaction time points of the source domain resources, the sample target domain interaction items include target domain resources and interaction time points of the target domain resources, the sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction item, and the sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction item; The method further comprises: For any source domain resource in the second sample source domain interaction data, determine a first target domain resource in the sample target domain interaction data, determine a source domain resource feature of the source domain resource and a target domain resource feature of the first target domain resource as a first positive sample pair, and the interaction time point of the first target domain resource is closest to the interaction time point of the source domain resource; The training of the resource recommendation model based on the recommendation prediction result and the sample label includes: The resource recommendation model is trained based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair.

13. The method according to claim 12, characterized in that The method further comprises: For any source domain resource in the second sample source domain interaction data, determine a second target domain resource in the sample target domain interaction data of any object other than the sample object, determine a source domain resource feature of the source domain resource and a target domain resource feature of the second target domain resource as a first negative sample pair, and the second target domain resource is any target domain resource in the sample target domain interaction data of the any object; The training of the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair includes: The resource recommendation model is trained based on the recommendation prediction result, the sample label, the similarity between the source domain resource feature and the target domain resource feature in the first positive sample pair, and the similarity between the source domain resource feature and the target domain resource feature in the first negative sample pair.

14. The method according to claim 11, characterized in that The sample source domain interaction items include source domain resources and entities to which the source domain resources belong, the sample target domain interaction items include target domain resources and entities to which the target domain resources belong, the sample source domain interaction features include source domain resource features of the source domain resources in each sample source domain interaction item, and the sample target domain interaction features include target domain resource features of the target domain resources in each sample target domain interaction item; The method further comprises: For any source domain resource in the second sample source domain interaction data, a third target domain resource is determined in the sample target domain interaction data, a source domain resource feature of the source domain resource and a target domain resource feature of the third target domain resource are determined as a second positive sample pair, and the entity to which the third target domain resource belongs is the same as the entity to which the source domain resource belongs; The training of the resource recommendation model based on the recommendation prediction result and the sample label includes: The resource recommendation model is trained based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the second positive sample pair.

15. The method according to claim 14, characterized in that The method further comprises: For any source domain resource in the second sample source domain interaction data, determine a fourth target domain resource, determine a source domain resource feature of the source domain resource and a target domain resource feature of the fourth target domain resource as a second negative sample pair, and the fourth target domain resource is any target domain resource different from the entity to which the source domain resource belongs; The training of the resource recommendation model based on the recommendation prediction result, the sample label, and the similarity between the source domain resource feature and the target domain resource feature in the second positive sample pair includes: The resource recommendation model is trained based on the recommendation prediction result, the sample label, the similarity between the source domain resource features and the target domain resource features in the second positive sample pair, and the similarity between the source domain resource features and the target domain resource features in the second negative sample pair.

16. A resource recommendation device, characterized in that: The device comprises: A first acquisition module, configured to acquire object data of an object and resource data of a target resource, wherein the target resource belongs to a target domain; A second acquisition module is provided for acquiring first source domain interaction data of the object and target domain interaction data of the object, wherein the first source domain interaction data includes a plurality of source domain interaction items, wherein the source domain interaction items refer to the interaction between the object and resources in the source domain, and the target domain interaction data includes a plurality of target domain interaction items, wherein the target domain interaction items refer to the interaction between the object and resources in the target domain, and the source domain and the target domain belong to different domains; a screening module, configured to determine, in the first source domain interaction data, at least one source domain interaction event associated with the target domain, wherein the at least one source domain interaction event associated with the target domain constitutes second source domain interaction data; A prediction module is used to determine a recommendation prediction result of the target resource based on the object data, the resource data, the second source domain interaction data and the target domain interaction data, wherein the recommendation prediction result is used to characterize the interaction between the object and the target resource if the target resource is recommended to the object.

17. A training device for a resource recommendation model, characterized in that: The device comprises: A first acquisition module, configured to acquire sample object data of a sample object, sample resource data of a sample resource, and a sample label, wherein the sample resource belongs to a target domain, and the sample label is used to indicate a real interaction between the sample object and the sample resource after the sample resource is recommended to the sample object; A second acquisition module is used to acquire first sample source domain interaction data of the sample object and sample target domain interaction data of the sample object, wherein the first sample source domain interaction data includes a plurality of sample source domain interaction events, wherein the sample source domain interaction events refer to the interaction between the sample object and resources in the source domain, and the sample target domain interaction data includes a plurality of sample target domain interaction events, wherein the sample target domain interaction events refer to the interaction between the sample object and resources in the target domain, and the source domain and the target domain belong to different domains; a screening module, configured to determine, in the first sample source domain interaction data, at least one sample source domain interaction event associated with the target domain, wherein the at least one sample source domain interaction event associated with the target domain constitutes the second sample source domain interaction data; A prediction module, configured to determine a recommendation prediction result of the sample resource based on the sample object data, the sample resource data, the second sample source domain interaction data, and the target domain sample interaction data through a resource recommendation model, wherein the recommendation prediction result is used to characterize a predicted interaction between the sample object and the sample resource if the sample resource is recommended to the sample object; A training module is used to train the resource recommendation model based on the recommendation prediction result and the sample label.

18. A computer device, characterized in that: The computer device includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource recommendation method as described in any one of claims 1 to 9, or to implement the operations performed by the resource recommendation model training method as described in any one of claims 10 to 15.

19. A computer-readable storage medium, characterized in that: At least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource recommendation method as described in any one of claims 1 to 9, or to implement the operations performed by the resource recommendation model training method as described in any one of claims 10 to 15.

20. A computer program product comprising a computer program, characterized in that The computer program is loaded and executed by a processor to implement the operations performed by the resource recommendation method as described in any one of claims 1 to 9, or to implement the operations performed by the resource recommendation model training method as described in any one of claims 10 to 15.

Citation Information

Cited By

  • Resource recommendation method and apparatus, resource recommendation model training method and apparatus, and device

    WO2026031903A1