A resource processing method, apparatus, electronic device, and storage medium

Abstract

The present application discloses a resource processing method, apparatus, electronic device, and storage medium. The method includes: inputting the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction to obtain a first resource feature and a first object feature; inputting the resource information and the object information into a second streaming feature extraction model for feature extraction to obtain a second resource feature and a second object feature, where the second streaming feature extraction model is obtained by training the model based on sample association features. Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determine the push metric data of the resource to be recommended and the target object, and perform resource recommendation based on the push metric data. This method can improve the effectiveness of the model in learning association features in online real-time training and improve the accuracy of rough ranking recall in online applications.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a resource processing method, apparatus, electronic device, and storage medium. Background Art

[0002] With the explosive growth of Internet information, the ability of a streaming recommendation system to train and update in real time based on newly added data is more suitable for big data scenarios. In actual applications, a streaming recommendation system also includes a rough ranking and recall stage and a fine ranking and recall stage. The range of resources to be recommended is narrowed through the rough ranking and recall stage, and then more accurate resource recall is performed through the fine ranking and recall stage.

[0003] In the prior art, the rough ranking and recall stage of a streaming recommendation system is limited by the rough ranking and recall strategy and the model architecture, and it is difficult to introduce the association information between the resources to be recommended and the target object. Therefore, it is difficult to discover the implicit interests of users, resulting in low accuracy of rough ranking and recall. Summary of the Invention

[0004] This application provides a resource processing method, apparatus, electronic device, and storage medium, which can improve the accuracy of rough ranking and recall.

[0005] On the one hand, this application provides a resource processing method, the method includes:

[0006] Input the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction, to obtain a first resource feature of the resource to be recommended and a first object feature of the target object; the first streaming feature extraction model is obtained by training the model based on the operation data of sample objects on sample resources;

[0007] Input the resource information and the object information into a second streaming feature extraction model for feature extraction, to obtain a second resource feature of the resource to be recommended and a second object feature of the target object; the second streaming feature extraction model is obtained by training the model based on sample association features; the sample association features represent the association information between the sample object and the sample resource;

[0008] Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determine the push index data of the resource to be recommended and the target object;

[0009] Push a target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push index data.

[0010] On the other hand, a resource processing apparatus is provided, the apparatus includes:

[0011] The first feature extraction module is configured to input the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction, so as to obtain a first resource feature of the resource to be recommended and a first object feature of the target object; the first streaming feature extraction model is trained based on the operation data of sample objects on sample resources.

[0012] The second feature extraction module is configured to input the resource information and the object information into a second streaming feature extraction model for feature extraction, so as to obtain a second resource feature of the resource to be recommended and a second object feature of the target object; the second streaming feature extraction model is trained based on sample association features; the sample association features represent the association information between the sample object and the sample resource.

[0013] The push metric determination module is configured to determine push metric data of the resource to be recommended and the target object based on the first resource feature, the second resource feature, the first object feature, and the second object feature.

[0014] The target recommended resource determination module is configured to push a target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push metric data.

[0015] On the other hand, an electronic device is provided, which includes a processor and a memory. At least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement a resource processing method as described above.

[0016] On the other hand, a computer-readable storage medium is provided, which includes a processor and a memory. At least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement a resource processing method as described above.

[0017] On the other hand, a computer program product is provided, which includes a computer program. When the computer program is executed by a processor, the resource processing method as described above is implemented.

[0018] A resource processing method, apparatus, electronic device, and storage medium provided by this application. The method includes: inputting the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction to obtain a first resource feature and a first object feature; inputting the resource information and the object information into a second streaming feature extraction model for feature extraction to obtain a second resource feature and a second object feature. The second streaming feature extraction model is trained based on sample association features, and the sample association features represent the association information between sample objects and sample resources. Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determine the push metric data of the resource to be recommended and the target object, and perform resource recommendation based on the push metric data. This method can train the second streaming feature extraction model online in real time and directly use the sample association features for model training, enabling the second streaming feature extraction model to directly learn the sample association features, thereby improving the effectiveness of the model in learning association features. When applied online, the second streaming feature extraction model can identify the potential association information between the resource to be recommended and the target object, thus improving the accuracy of rough ranking recall. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 Schematic diagram of an application scenario of a resource processing method provided by an embodiment of this application;

[0021] Figure 2 Flowchart of a resource processing method provided by an embodiment of this application;

[0022] Figure 3 Flowchart of a method for feature extraction based on multiple association feature extraction models in a resource processing method provided by an embodiment of this application;

[0023] Figure 4 Schematic diagram of the connection structure between a first streaming feature extraction model and an association feature extraction model in a resource processing method provided by an embodiment of this application;

[0024] Figure 5 Schematic diagram of the connection structure between a first streaming feature extraction model and multiple association feature extraction models in a resource processing method provided by an embodiment of this application;

[0025] Figure 6Schematic diagram of the structures of a multi-layer resource feature extraction layer and a multi-layer object feature extraction layer in a resource processing method provided by an embodiment of the present application;

[0026] Figure 7 Flowchart of a method for determining push index data in a resource processing method provided by an embodiment of the present application;

[0027] Figure 8 Flowchart of a method for training a second streaming feature extraction model in a resource processing method provided by an embodiment of the present application;

[0028] Figure 9 Flowchart of a method for training a model based on a first sample associated sub-feature in a resource processing method provided by an embodiment of the present application;

[0029] Figure 10 Flowchart of a method for training a model based on a second sample associated sub-feature in a resource processing method provided by an embodiment of the present application;

[0030] Figure 11 Schematic diagram of offline scoring of a fine-rank model and feature distillation of a two-tower model based on online scoring provided by an embodiment of the present application;

[0031] Figure 12 Schematic diagram of online scoring of a fine-rank model and feature distillation of a two-tower model based on online scoring provided by an embodiment of the present application;

[0032] Figure 13 Schematic diagram of joint training of a fine-rank model and a two-tower model provided by an embodiment of the present application;

[0033] Figure 14 Schematic diagram of the structure of a model for executing a resource processing method in a resource processing method provided by an embodiment of the present application;

[0034] Figure 15 Schematic diagram of the structure of a resource processing device provided by an embodiment of the present application;

[0035] Figure 16 Schematic diagram of the hardware structure of a device for implementing the method provided by an embodiment of the present application provided by an embodiment of the present application. Detailed implementation manners

[0036] To make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0037] In the description of the present application, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. Moreover, terms such as "first" and "second" are applicable to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order different from those illustrated or described here.

[0038] It can be understood that in the specific implementation manners of the present application, when it comes to data related to user information, etc., when the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0039] First, the following explanations are made for the relevant terms involved in the embodiments of the present application:

[0040] The Hadamard product is a type of matrix operation. If A = (aij) and B = (bij) are two matrices of the same order, and if cij = aij × bij, then the matrix C = (cij) is called the Hadamard product of A and B.

[0041] Please refer to Figure 1 , which shows a schematic diagram of an application scenario of a resource processing method provided by an embodiment of the present application. The application scenario includes a client 110 and a server 120. The client 110 sends the resource to be recommended and the target object to the server 120. The server 120 performs rough ranking and recall based on a first streaming feature extraction model and a second streaming feature extraction model to obtain the target recommended resource. The server 120 sends the target recommended resource to the client 110. The server 120 can also online train the second streaming feature extraction model based on the sample correlation features.

[0042] In the embodiments of the present application, the client 110 includes but is not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc.

[0043] In the embodiments of the present application, the server 120 may include an independently operating server, or a distributed server, or a server cluster composed of multiple servers. The server 120 may include a network communication unit, a processor, a memory, and so on.

[0044] Please refer to Figure 2, which shows a resource processing method applicable to the server side. The method includes:

[0045] S210. Input the resource information of the resource to be recommended and the object information of the target object into the first streaming feature extraction model for feature extraction, to obtain the first resource feature of the resource to be recommended and the first object feature of the target object; the first streaming feature extraction model is trained based on the operation data of the sample object on the sample resource;

[0046] In some embodiments, the first streaming feature extraction model is trained online in real time, and is a model that separately extracts features from the resource information of the resource to be recommended and the object information of the target object. The first streaming feature extraction model can only separately extract the single linear features corresponding to the resource information and the object information respectively, and cannot extract the compound non-linear features corresponding to the associated information between the resource information and the object information.

[0047] The resource information is linear information directly obtained based on the resource identifier of the resource to be recommended, and the object information is linear information directly obtained based on the object identifier of the target object. The potential association between the resource information and the object information belongs to non-linear information and cannot be directly obtained through the object identifier or the resource identifier. For example, the object identifier is the identity document (id) of the user who uniquely identifies the video application, and the resource identifier is the id of the video in the video application. Then, through the user's id, information such as the user's age, gender, and preferences can be determined, which is the object information obtained based on the object identifier. Through the video's id, information such as the video's category, title, and introduction can be determined, which is the resource information obtained based on the resource identifier. Whether the user is likely to follow the video author belongs to the information obtained through the potential association between the user's id and the video's id, and cannot be used as the resource information or the object information.

[0048] S220. Input the resource information and the object information into the second streaming feature extraction model for feature extraction, to obtain the second resource feature of the resource to be recommended and the second object feature of the target object; the second streaming feature extraction model is trained based on the sample association features; the sample association features represent the association information between the sample object and the sample resource;

[0049] In some embodiments, the second streaming feature extraction model can be trained online in real time. When performing feature extraction on the resource information of the resource to be recommended and the object information of the target object respectively, the second resource feature and the second object feature containing the potential association information between the resource information and the object information are extracted. The second streaming feature extraction model can be obtained by training based on the sample association features. Thus, when the model input only contains resource information and object information, the second streaming feature extraction model can have the function of identifying the potential association information between the resource information and the object information.

[0050] In some embodiments, see Figure 3 , the sample association features include multiple sample association sub-features, and the second streaming feature extraction model includes association feature extraction models corresponding to the multiple sample association sub-features respectively;

[0051] Inputting the resource information and the object information into the second streaming feature extraction model for feature extraction to obtain the second resource feature of the resource to be recommended and the second object feature of the target object includes:

[0052] S310. Input the resource information and the object information into each association feature extraction model for feature extraction respectively to obtain multiple associated resource features of the resource to be recommended and multiple associated object features of the target object;

[0053] S320. Perform feature splicing processing on the multiple associated resource features to obtain the second resource feature;

[0054] S330. Perform feature splicing processing on the multiple object resource features to obtain the second object feature.

[0055] In some embodiments, the sample association sub-features can be obtained based on different feature extraction methods. For example, they can be obtained by directly performing feature extraction based on the association information between the sample resource and the sample object, or by performing attention learning on the sample resource information of the sample resource and the sample object information of the sample object. Different sample association sub-features have association information with different dimensions or different contents.

[0056] When the sample association features include multiple sample association sub-features, the second streaming feature extraction model can include multiple association feature extraction models. Different association feature extraction models are obtained by training based on different sample association sub-features.

[0057] The first streaming feature extraction model and the second streaming feature extraction model are in a parallel relationship. As Figure 4 shown, when there is only one association feature extraction model in the second streaming feature extraction model, the first streaming feature extraction model and this association feature extraction model are in parallel. As Figure 5As shown, in the case where there are multiple associated feature extraction models in the second streaming feature extraction model, the first streaming feature extraction model is in parallel connection with each of the multiple associated feature extraction models, and the multiple associated feature extraction models are also in parallel connection with each other.

[0058] Input the resource information and object information into each of the associated feature extraction models for feature extraction, and multiple associated resource features of the resource to be recommended and multiple associated object features of the target object can be obtained. Since the types of sample associated sub-features used in the training of each associated feature extraction model are different, different types of sample associated sub-features will focus on different associated information. Therefore, there are differences in the types of associated information corresponding to each pair of associated resource features, and there are also differences in the types of associated information corresponding to each pair of associated object features.

[0059] Perform feature splicing processing on the multiple associated resource features to obtain the second resource feature. Perform feature splicing processing on the multiple object resource features to obtain the second object feature. The second resource feature and the second object feature are features containing different types of potential associated information.

[0060] By splicing the features output by each associated feature extraction model, different types of potential associated information can be obtained, thereby improving the comprehensiveness of the second resource feature and the second object feature.

[0061] In some embodiments, the second streaming feature extraction model includes multiple layers of resource feature extraction layers and multiple layers of object feature extraction layers; the number of layers of the multiple layers of resource feature extraction layers is the same as the number of layers of the multiple layers of object feature extraction layers;

[0062] Input the resource information and object information into the second streaming feature extraction model for feature extraction to obtain the second resource feature of the resource to be recommended and the second object feature of the target object, including:

[0063] Based on the multiple layers of resource feature extraction layers and the resource information for feature extraction, obtain the second resource feature;

[0064] Based on the multiple layers of object feature extraction layers and the object information for feature extraction, obtain the second object feature.

[0065] In some embodiments, the second streaming feature extraction model can be obtained by stacking multiple layers of resource feature extraction layers and multiple layers of object feature extraction layers, thereby increasing the number of network layers of the second streaming feature extraction model.

[0066] The multi-layer resource feature extraction layers are arranged in sequence. The resource information is input into the multi-layer resource feature extraction layers for feature extraction in sequence. In the multi-layer resource feature extraction layers, the output information of the upper-layer resource feature extraction layer is the input information of the lower-layer resource feature extraction layer, and the output information of the last-layer resource feature extraction layer is used as the second resource feature obtained.

[0067] The multi-layer object feature extraction layers are arranged in sequence. The object information is input into the multi-layer object feature extraction layers for feature extraction in sequence. In the multi-layer object feature extraction layers, the output information of the upper-layer object feature extraction layer is the input information of the lower-layer object feature extraction layer, and the output information of the last-layer object feature extraction layer is used as the second object feature obtained.

[0068] Please refer to Figure 6 , such as Figure 6 shown in the schematic diagram of superimposing the second resource feature extraction layer and the second object feature extraction layer on the first resource feature extraction layer and the first object feature extraction layer respectively. The first resource feature extraction layer is cascaded with the second resource feature extraction layer, and the first object feature extraction layer is cascaded with the second object feature extraction layer. The information output by the second resource feature extraction layer is used as the second resource feature, and the information output by the second object feature extraction layer is used as the second object feature.

[0069] Through the multi-layer resource feature extraction layers and the multi-layer object feature extraction layers, the network depth of the second streaming feature extraction model is deepened, so as to obtain the second resource feature and the second object feature that can better express potential association information, and the effectiveness of the second resource feature and the second object feature is improved.

[0070] S230. Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determine the push index data of the resource to be recommended and the target object;

[0071] In some embodiments, please refer to Figure 7 , based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determining the push index data of the resource to be recommended and the target object includes:

[0072] S710. Perform feature splicing processing on the first resource feature and the second resource feature to obtain the target resource feature;

[0073] S720. Perform feature splicing processing on the first object feature and the second object feature to obtain the target object feature;

[0074] S730. Based on the target resource feature and the target object feature, perform feature cross-processing to determine the push index data.

[0075] In some embodiments, for the feature splicing process of the first resource feature and the second resource feature, the feature splicing can be performed through the concat function to obtain the target resource feature, which includes both the feature information corresponding to the resource to be recommended and the feature information corresponding to the potential association information between the resource to be recommended and the target object. For the feature splicing process of the first object feature and the second object feature, the feature splicing can be performed through the concat function to obtain the target object feature, which includes both the feature information corresponding to the target object and the feature information corresponding to the potential association information between the resource to be recommended and the target object.

[0076] When performing feature cross-processing based on the target resource feature and the target object feature, the feature cross-processing can be the vector inner product of the target resource feature and the target object feature to determine the push metric data. The push metric data can include the click-through rate (CTR) and the conversion rate (CVR). Therefore, based on the push metric data, it can be determined whether the target object will click on the video to be recommended or whether the target object will follow the video publisher through the video to be recommended. Thus, based on the push metric data, the target recommended resource can be determined from the resources to be recommended.

[0077] By splicing the features output by the first streaming feature extraction model and the features output by the second streaming feature extraction model, the non-linear features and the linear features can be fused into the target resource feature and the target object feature, thereby improving the accuracy of the push metric data through the fused target resource feature and target object feature.

[0078] S240. Push the target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push metric data.

[0079] In some embodiments, the target recommended resource can be the rough ranking recall result in resource recommendation. Based on the target recommended resource, the determination of the recommended object can be performed again to obtain the fine ranking recall result. The number of resources in the fine ranking recall result is less than that in the rough ranking recall result, and the accuracy of the fine ranking recall result is greater than that of the rough ranking recall result.

[0080] Push the target recommended resource to the target object, or push the fine ranking recall result corresponding to the target recommended resource to the target object.

[0081] In some embodiments, please refer to Figure 8 , this method is a method for training the second streaming feature extraction model, and this method further includes:

[0082] S810. Determine the sample association features based on a preset resource recommendation model; the preset resource recommendation model is obtained by training a model based on the sample resource information of the sample resources, the sample object information of the sample objects, and the association information between the sample objects and the sample resources;

[0083] S820. Input the sample resource information and the sample object information into the model to be trained for feature extraction, and obtain the training resource features corresponding to the sample resources and the training object features corresponding to the sample objects;

[0084] S830. Perform feature association processing between the training resource features and the training object features to obtain training association features;

[0085] S840. Based on the sample association features and the training association features, train the model to be trained to obtain a second streaming feature extraction model.

[0086] In some embodiments, the preset resource recommendation model can be a model for performing fine-rank recall. The preset resource recommendation model is obtained by training a model based on the sample resource information of the sample resources, the sample object information of the sample objects, and the association information between the sample objects and the sample resources. The preset resource recommendation model can extract the composite features corresponding to the association information between the sample resource information and the sample object information, and then determine the push metric data based on the composite features, so as to obtain the fine-rank recall result. Therefore, the sample association features can be obtained from the preset resource recommendation model. The sample association features are also composite non-linear features, which can represent the association information between the sample resources and the sample objects.

[0087] Inputting the sample resource information and the sample object information into the model to be trained for feature extraction can obtain the training resource features corresponding to the sample resources and the training object features corresponding to the sample objects. The model to be trained is a model for separately performing feature extraction on the sample resource information of the sample resources and the sample object information of the sample objects. The model to be trained includes a sample resource feature extraction layer and a sample object feature extraction layer. The sample resource feature extraction layer is used to perform feature extraction on the sample resource information to obtain the training resource features, and the sample object feature extraction layer is used to perform feature extraction on the sample object information to obtain the training object features.

[0088] Performing feature association processing between the training resource features and the training object features can obtain training association features. When performing feature association processing, the first feature matrix corresponding to the training resource features and the second feature matrix corresponding to the training object features can be determined, and the product between the first feature matrix and the second training feature matrix can be calculated, so as to obtain the Hadamard product between the training resource features and the training object features. And use this Hadamard product as the training association features.

[0089] Based on the feature similarity between the sample correlation features and the training correlation features, loss information can be determined. Based on this loss information, the model to be trained is trained to obtain the second streaming feature extraction model. The feature similarity can be the cosine distance. In the case of a larger feature similarity, the training correlation features are more similar to the sample correlation features. Therefore, after model training, the potential correlation information between the to-be-recommended resource and the target object can be expressed in the second resource features and the second object features output by the second streaming feature extraction model.

[0090] By using the sample correlation features obtained based on the preset resource recommendation model to train the model to be trained, the second streaming feature extraction model is obtained, enabling the second streaming feature extraction model to directly learn the sample correlation features, thereby improving the effectiveness of the model in learning correlation features, enabling the acquisition of correlation features in a model that can only obtain linear features, thus improving the applicability of the second streaming feature extraction model and the accuracy of resource recall in the rough ranking stage.

[0091] In some embodiments, please refer to Figure 9 , the sample correlation features include the first sample correlation sub-features, the preset resource recommendation model includes a feature extraction layer, and based on the preset resource recommendation model, determining the sample correlation features includes:

[0092] S910. Input the correlation information into the feature extraction layer for feature extraction to obtain the first sample correlation sub-features;

[0093] The second streaming feature extraction model includes the first correlation feature extraction model. By using the sample correlation features and the training correlation features to train the model to be trained, obtaining the second streaming feature extraction model includes:

[0094] S920. Based on the first sample correlation sub-features and the training correlation features, determine the first loss information;

[0095] S930. Based on the first loss information, train the model to be trained to obtain the first correlation feature extraction model.

[0096] In some embodiments, the correlation information between the sample resource and the sample object can be obtained by performing feature preprocessing on the sample resource information and the sample object information. The feature preprocessing can perform non-linear feature synthesis on the sample resource information and the sample object information, such as inner product, division, squaring, etc., thereby obtaining the correlation information. Among them, the sample resource information and the sample object information can be vector representations.

[0097] The association information between the sample resource and the sample object can be directly input into the feature extraction layer of the preset resource recommendation model to directly extract features from the association information, thereby obtaining the first sample association sub-feature. Based on the feature similarity between the first sample association sub-feature and the training association feature, the first loss information is determined, and then based on the first loss information, the model to be trained is trained to obtain the first association feature extraction model. The first association feature extraction model is a model that is biased towards directly identifying the non-linear features between the resource to be recommended and the target object.

[0098] The first association feature extraction model can be used as the second streaming feature extraction model, or other association feature extraction models can be trained through different sample association sub-features, and all the trained association feature extraction models are used as the second streaming feature extraction model.

[0099] By obtaining the first sample association sub-feature through the association information between the sample resource and the sample object and training the model to be trained to obtain the first association feature extraction model, it is possible to obtain association features in a model that can only obtain linear features, thereby improving the applicability of the second streaming feature extraction model and improving the accuracy of resource recall in the rough ranking stage.

[0100] In some embodiments, please refer to Figure 10 , the sample association feature includes a second sample association sub-feature, and the preset resource recommendation model includes an attention learning layer. Based on the preset resource recommendation model, determining the sample association feature includes:

[0101] S1010. Input the sample resource information and the sample object information into the attention learning layer for attention learning to obtain the second sample association sub-feature;

[0102] The second streaming feature extraction model includes a second association feature extraction model. Based on the sample association feature and the training association feature, training the model to be trained to obtain the second streaming feature extraction model includes:

[0103] S1020. Based on the second sample association sub-feature and the training association feature, determine the second loss information;

[0104] S1030. Based on the second loss information, train the model to be trained to obtain the second association feature extraction model.

[0105] In some embodiments, the sample resource information input into the preset resource recommendation model may include a resource author identifier, a resource tag, etc., and the sample object information may include a sample object identifier, portrait information, a user click list, a user play duration list, a user interaction list, a user negative feedback list, etc.

[0106] Inputting the sample resource information and the sample object information into the attention learning layer for attention learning can enable the sample resource information and the sample object information to cross each other, thereby obtaining the second sample correlation sub-feature. Based on the feature similarity between the second sample correlation sub-feature and the training correlation feature, the second loss information is determined, and then based on the second loss information, the model to be trained is trained to obtain the second correlation feature extraction model. The second correlation feature extraction model is a model that is biased towards identifying the non-linear feature between the resource to be recommended and the target object by identifying the attention cross feature between the resource to be recommended and the target object.

[0107] By obtaining the second sample correlation sub-feature through the cross information in the attention learning layer and training the model to be trained to obtain the second correlation feature extraction model, it is possible to obtain correlation features in a model that can only obtain linear features, thereby improving the applicability of the second streaming feature extraction model and improving the accuracy of resource recall in the rough ranking stage.

[0108] In some embodiments, the resource processing method can be applied to the rough ranking and recall stage of a real-time recommendation system. Both the first streaming feature extraction model and the second streaming feature extraction model can use a two-tower model. The two towers in the two-tower model can respectively extract features related to resources and features related to objects.

[0109] In the prior art, as Figure 11 shown, a fixed version of the fine-ranking model can be used to perform offline scoring on the sample data of one day. The two-tower model can calculate the loss information between the push metric data and the offline scoring, and can distill the offline scoring of the fine-ranking model, thereby learning the cross features. However, since the offline scoring by the fine-ranking model takes a long time and cannot be trained in a streaming manner, and in this method, the two-tower model learns the cross features during the feature processing of the fine-ranking model by distilling the output of the fine-ranking model, and the way of learning the cross features is relatively indirect.

[0110] In the prior art, as Figure 12 shown, the online scoring output by the online fine-ranking model can be used, and the online scoring is stored in the sample data. The two-tower model can be trained through the loss information between the push metric data and the sample data, and can distill the online scoring of the online fine-ranking model, thereby learning the cross features. However, the online scoring output by the online fine-ranking model is not stable. After streaming training, due to the fast change of the states of each video in the video application, the distribution of the online scoring also changes quickly. And in the online scenario, because there are usually experiments online and not all traffic uses the same fine-ranking model, only some samples may be available for the distilled fine-ranking scoring, making it difficult to learn accurate cross features.

[0111] In the prior art, asFigure 13 As shown in the figure, the dual-tower push metric data output by the dual-tower model and the refined ranking push metric data output by the refined ranking model are used to jointly train the dual-tower model and the refined ranking model. What the dual-tower model distills is the training score of the refined ranking model. This solution also learns the output of the refined ranking model rather than the direct cross features, resulting in relatively indirect learning of the cross features.

[0112] As Figure 14 As shown in the figure is a schematic structural diagram of a model for executing a resource processing method. In the resource processing method of the embodiment of the present application, the resource to be recommended and the target object are obtained in real time. The resource to be recommended can be a video resource, an image resource, an audio resource, etc., and the target object can be a user. The resource information of the resource to be recommended and the object information of the target object are input into the first streaming feature extraction model for feature extraction. The first streaming feature extraction model is a dual-tower model, and the structure of the dual towers is composed of an initial resource feature extraction layer and an initial object feature extraction layer. By inputting the resource information into the initial resource feature extraction layer for feature extraction, the first resource feature of the resource to be recommended can be obtained. By inputting the object information into the initial object feature extraction layer for feature extraction, the first object feature of the target object can be obtained. The first streaming feature extraction model can only extract single linear features and cannot extract composite non-linear features.

[0113] The resource information and the object information are input into the second streaming feature extraction model for feature extraction. The second streaming feature extraction model is a dual-tower model, and the structure of the dual towers is composed of a resource feature extraction layer and an object feature extraction layer. Among them, both the resource feature extraction layer and the object feature extraction layer can have multiple layers, and the number of layers of the resource feature extraction layer is the same as the number of layers of the object feature extraction layer. By inputting the resource information into the resource feature extraction layer for feature extraction, the second resource feature of the resource to be recommended can be obtained. By inputting the object information into the resource feature extraction layer for feature extraction, the second object feature of the target object can be obtained.

[0114] Among them, the second streaming feature extraction model is obtained by training the model based on the sample correlation features. The sample correlation features represent the correlation information between the sample object and the sample resource, that is, the cross features between the sample object and the sample resource, and can be obtained from a preset resource recommendation model. The preset resource recommendation model is a resource recommendation model applied to the refined ranking and recall stage. The preset resource recommendation model can process composite non-linear features and single linear features. That is to say, the preset resource recommendation model can extract features from the resource information, the object information, and the correlation information between the resource to be recommended and the target object, and perform resource recommendation based on the extracted features.

[0115] During model training, the training resource features and training sample features output by the model to be trained can be subjected to feature association processing to obtain training association features. Based on the loss information obtained from the training association features and sample association features, the model to be trained is trained to obtain a second streaming feature extraction model. By ensuring that the similarity between the training association features and the sample association features meets a preset similarity threshold, the training association features can approximately represent the sample association features, enabling the second streaming feature extraction model to learn cross features.

[0116] The sample association features can include multiple sample association sub-features. The feature extraction layer of the second streaming feature extraction model can directly obtain the first sample association sub-feature from the association information between the sample object and the sample resource, or can perform attention learning on the sample resource information and the sample object information through the attention learning layer of the second streaming feature extraction model to obtain the second sample association sub-feature related to attention learning.

[0117] Correspondingly, the second streaming feature extraction model can include association feature extraction models corresponding to different sample association sub-features. Different association feature extraction models are obtained by training the model based on different sample association sub-features.

[0118] By performing feature splicing processing on the first resource feature and the second resource feature, the target resource feature can be obtained. By performing feature splicing processing on the first object feature and the second object feature, the target object feature can be obtained. By performing feature cross-processing between the target resource feature and the target object feature, the push metric data of the resource to be recommended and the target object is determined. The push metric data can include CTR and CVR.

[0119] Based on the push metric data, the target recommended resource is determined from the resources to be recommended. The target recommended resource is the rough ranking recall result. Based on the target recommended resource and a preset resource recommendation model, the refined ranking recall result can be obtained for the recommended resource. The refined ranking recall result is sent to the target object.

[0120] An embodiment of the present application provides a resource processing method, which includes: inputting the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction to obtain the first resource feature of the resource to be recommended and the first object feature of the target object; inputting the resource information and the object information into a second streaming feature extraction model for feature extraction to obtain the second resource feature of the resource to be recommended and the second object feature of the target object. The second streaming feature extraction model is trained based on sample association features, and the sample association features represent the association information between the sample object and the sample resource. Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determine the push metric data of the resource to be recommended and the target object, and perform resource recommendation based on the push metric data. This method can train the second streaming feature extraction model online in real time and directly use the sample association features for model training, so that the second streaming feature extraction model can directly learn the sample association features, thereby improving the effectiveness of the model in learning association features. When applied online, the second streaming feature extraction model can identify the potential association information between the resource to be recommended and the target object, thereby improving the accuracy of rough ranking recall.

[0121] An embodiment of the present application also provides a resource processing device, please refer to Figure 15 , the device includes:

[0122] A first feature extraction module 1510, configured to input the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction to obtain the first resource feature of the resource to be recommended and the first object feature of the target object; the first streaming feature extraction model is trained based on the operation data of the sample object on the sample resource;

[0123] A second feature extraction module 1520, configured to input the resource information and the object information into a second streaming feature extraction model for feature extraction to obtain the second resource feature of the resource to be recommended and the second object feature of the target object; the second streaming feature extraction model is trained based on sample association features; the sample association features represent the association information between the sample object and the sample resource;

[0124] A push metric determination module 1530, configured to determine the push metric data of the resource to be recommended and the target object based on the first resource feature, the second resource feature, the first object feature, and the second object feature;

[0125] A target recommended resource determination module 1540, configured to push a target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push metric data.

[0126] In some embodiments, the sample association features include multiple sample association sub - features, the second streaming feature extraction model includes association feature extraction models corresponding to the multiple sample association sub - features respectively, and the second feature extraction module includes:

[0127] An association feature extraction unit, configured to input the resource information and the object information into each association feature extraction model respectively for feature extraction, so as to obtain multiple association resource features of the resource to be recommended and multiple association object features of the target object;

[0128] A first association feature splicing unit, configured to perform feature splicing processing on the multiple association resource features to obtain a second resource feature;

[0129] A second association feature splicing unit, configured to perform feature splicing processing on the multiple object resource features to obtain a second object feature.

[0130] In some embodiments, the second streaming feature extraction model includes multiple layers of resource feature extraction layers and multiple layers of object feature extraction layers, the number of layers of the multiple layers of resource feature extraction layers is the same as the number of layers of the multiple layers of object feature extraction layers, and the second feature extraction module includes:

[0131] A second resource feature extraction unit, configured to perform feature extraction based on the multiple layers of resource feature extraction layers and the resource information to obtain a second resource feature;

[0132] A second object feature extraction unit, configured to perform feature extraction based on the multiple layers of object feature extraction layers and the object information to obtain a second object feature.

[0133] In some embodiments, the push metric determination module includes:

[0134] A resource feature splicing unit, configured to perform feature splicing processing on the first resource feature and the second resource feature to obtain a target resource feature;

[0135] An object feature splicing unit, configured to perform feature splicing processing on the first object feature and the second object feature to obtain a target object feature;

[0136] A push metric data determination unit, configured to perform feature cross - processing based on the target resource feature and the target object feature to determine push metric data.

[0137] In some embodiments, the apparatus further includes:

[0138] A sample association feature determination module, configured to determine sample association features based on a preset resource recommendation model; the preset resource recommendation model is obtained by model training based on the sample resource information of sample resources, the sample object information of sample objects, and the association information between sample objects and sample resources;

[0139] A training feature extraction module, configured to input sample resource information and sample object information into a model to be trained for feature extraction, so as to obtain training resource features corresponding to the sample resources and training object features corresponding to the sample objects;

[0140] A feature association processing module, configured to perform feature association processing between the training resource features and the training object features to obtain training association features;

[0141] A model training module, configured to perform model training on the model to be trained based on the sample association features and the training association features to obtain a second streaming feature extraction model.

[0142] In some embodiments, the sample association features include first sample association sub-features, the preset resource recommendation model includes a feature extraction layer, and the sample association feature determination module includes:

[0143] A first sample association sub-feature determination module, configured to input association information into the feature extraction layer for feature extraction to obtain first sample association sub-features;

[0144] The second streaming feature extraction model includes a first association feature extraction model, and the model training module includes:

[0145] A first loss information determination unit, configured to determine first loss information based on the first sample association sub-features and the training association features;

[0146] A first model training unit, configured to perform model training on the model to be trained based on the first loss information to obtain a first association feature extraction model.

[0147] In some embodiments, the sample association features include second sample association sub-features, the preset resource recommendation model includes an attention learning layer, and the sample association feature determination module includes:

[0148] A second sample association sub-feature determination module, configured to input the sample resource information, the sample object information, and the association information into the attention learning layer for attention learning to obtain second sample association sub-features;

[0149] The second streaming feature extraction model includes a second association feature extraction model, and the model training module includes:

[0150] A second loss information determination unit, configured to determine second loss information based on the second sample association sub-features and the training association features;

[0151] A second model training unit, configured to perform model training on the model to be trained based on the second loss information to obtain a second association feature extraction model.

[0152] The device provided in the above embodiments can execute the methods provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects for executing the methods. For technical details not described in detail in the above embodiments, reference may be made to a resource processing method provided in any embodiment of the present application.

[0153] This embodiment also provides a computer-readable storage medium, in which computer-executable instructions are stored, and the computer-executable instructions are loaded and executed by a processor to execute the above-mentioned resource processing method of this embodiment.

[0154] This embodiment also provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the various optional implementation manners of the above-mentioned resource processing.

[0155] This embodiment also provides an electronic device, which includes a processor and a memory. Among them, the memory stores a computer program, and the computer program is adapted to be loaded and executed by the processor to execute the above-mentioned resource processing method of this embodiment.

[0156] The device may be a computer terminal, a mobile terminal or a server, and the device may also participate in constituting the device or system provided in the embodiments of the present application. As Figure 16 shown, the server 16 may include one or more (shown as 1602a, 1602b,..., 1602n in the figure) processors 1602 (the processor 1602 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1604 for storing data, and a transmission device 1606 for communication functions. In addition, it may further include: an input / output interface (I / O interface), a network interface. Those of ordinary skill in the art can understand that Figure 16 the structure shown is only schematic, and it does not limit the structure of the above-mentioned electronic device. For example, the server 16 may further include more or fewer components than Figure 16 shown, or have a different configuration from Figure 16 shown.

[0157] It should be noted that the above one or more processors 1602 and / or other data processing circuits are generally referred to as "data processing circuits" in this article. The data processing circuit may be embodied in whole or in part as software, hardware, firmware or any other combination. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the server 16.

[0158] The memory 1604 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of the present application. The processor 1602 executes various functional applications and data processing by running the software programs and modules stored in the memory 1604, that is, implements the above-mentioned method for generating a temporal behavior capture frame based on a self-attention network. The memory 1604 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 1604 may further include a memory remotely disposed relative to the processor 1602, and these remote memories can be connected to the server 16 through a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0159] The transmission device 1606 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the server 16. In one instance, the transmission device 1606 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station and thus communicate with the Internet.

[0160] This specification provides the method operation steps as described in the embodiments or flowcharts, but may include more or fewer operation steps based on routine or non-creative labor. The steps and sequences listed in the embodiments are only one way among many execution sequences of the steps, and do not represent the only execution sequence. When the actual system or interrupt product is executed, it can be executed in the order of the method shown in the embodiments or the drawings or executed in parallel (for example, in a parallel processor or multi-threaded processing environment).

[0161] The structure shown in this embodiment is only a partial structure related to the solution of the present application, and does not constitute a limitation on the device to which the solution of the present application is applied. The specific device may include more or fewer components than those shown, or combine certain components, or have a different arrangement of components. It should be understood that the methods, devices, etc. disclosed in this embodiment can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, indirect coupling or communication connection of device or unit modules.

[0162] Based on such understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0163] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this specification can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present application.

[0164] As described above, the above embodiments are only used to illustrate the technical solution of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A resource processing method, characterized in that, The method includes: Inputting the resource information of the resource to be recommended and the object information of the target object into a first streaming feature extraction model for feature extraction to obtain a first resource feature of the resource to be recommended and a first object feature of the target object; the first streaming feature extraction model is trained based on the operation data of sample objects on sample resources; Inputting the resource information and the object information into a second streaming feature extraction model for feature extraction to obtain a second resource feature of the resource to be recommended and a second object feature of the target object; the second streaming feature extraction model is trained based on sample association features; the sample association features represent the association information between the sample object and the sample resource; Based on the first resource feature, the second resource feature, the first object feature, and the second object feature, determining the push index data of the resource to be recommended and the target object; Pushing a target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push index data.

2. The resource processing method according to claim 1, wherein, The sample association features include multiple sample association sub-features, and the second streaming feature extraction model includes association feature extraction models corresponding to the multiple sample association sub-features respectively; The step of inputting the resource information and the object information into the second streaming feature extraction model for feature extraction to obtain a second resource feature of the resource to be recommended and a second object feature of the target object includes: Inputting the resource information and the object information into each association feature extraction model for feature extraction to obtain multiple associated resource features of the resource to be recommended and multiple associated object features of the target object; Performing feature splicing processing on the multiple associated resource features to obtain the second resource feature; Performing feature splicing processing on the multiple associated object features to obtain the second object feature.

3. The resource processing method according to claim 1, wherein The second streaming feature extraction model includes multiple layers of resource feature extraction layers and multiple layers of object feature extraction layers; the number of layers of the multiple layers of resource feature extraction layers is the same as the number of layers of the multiple layers of object feature extraction layers; The step of inputting the resource information and the object information into the second streaming feature extraction model for feature extraction to obtain a second resource feature of the resource to be recommended and a second object feature of the target object includes: Performing feature extraction based on the multiple layers of resource feature extraction layers and the resource information to obtain the second resource feature; Performing feature extraction based on the multiple layers of object feature extraction layers and the object information to obtain the second object feature.

4. The resource processing method according to claim 1, wherein The step of determining the push index data of the resource to be recommended and the target object based on the first resource feature, the second resource feature, the first object feature, and the second object feature includes: Performing feature splicing processing on the first resource feature and the second resource feature to obtain a target resource feature; Performing feature splicing processing on the first object feature and the second object feature to obtain a target object feature; Based on the target resource features and the target object features, perform feature cross - processing to determine the push metric data.

5. The resource processing method according to claim 1, wherein The method further includes: Based on a preset resource recommendation model, determine the sample association features; the preset resource recommendation model is obtained by model training based on the sample resource information of the sample resources, the sample object information of the sample objects, and the association information between the sample objects and the sample resources; Input the sample resource information and the sample object information into the model to be trained for feature extraction, obtaining the training resource features corresponding to the sample resources and the training object features corresponding to the sample objects; Perform feature association processing between the training resource features and the training object features to obtain training association features; Based on the sample association features and the training association features, perform model training on the model to be trained to obtain the second streaming feature extraction model.

6. The resource processing method according to claim 5, wherein The sample association features include first sample association sub - features, and the preset resource recommendation model includes a feature extraction layer. Based on the preset resource recommendation model, determining the sample association features includes: Input the association information into the feature extraction layer for feature extraction to obtain the first sample association sub - features; The second streaming feature extraction model includes a first association feature extraction model. Based on the sample association features and the training association features, performing model training on the model to be trained to obtain the second streaming feature extraction model includes: Based on the first sample association sub - features and the training association features, determine the first loss information; Based on the first loss information, perform model training on the model to be trained to obtain the first association feature extraction model.

7. The resource processing method according to claim 5, wherein The sample association features include second sample association sub - features, and the preset resource recommendation model includes an attention learning layer. Based on the preset resource recommendation model, determining the sample association features includes: Input the sample resource information, the sample object information, and the association information into the attention learning layer for attention learning to obtain the second sample association sub - features; The second streaming feature extraction model includes a second association feature extraction model. Based on the sample association features and the training association features, performing model training on the model to be trained to obtain the second streaming feature extraction model includes: Based on the second sample association sub - features and the training association features, determine the second loss information; Based on the second loss information, perform model training on the model to be trained to obtain the second association feature extraction model.

8. A resource processing device, characterized in that, The apparatus includes: A first feature extraction module, configured to input the resource information of the resource to be recommended and the object information of the target object into the first streaming feature extraction model for feature extraction, obtaining the first resource features of the resource to be recommended and the first object features of the target object; the first streaming feature extraction model is obtained by model training based on the operation data of the sample objects on the sample resources. A second feature extraction module, configured to input the resource information and the object information into a second streaming feature extraction model for feature extraction, so as to obtain a second resource feature of the resource to be recommended and a second object feature of the target object; the second streaming feature extraction model is obtained by training a model based on sample association features; the sample association features represent the association information between the sample object and the sample resource; A push metric determination module, configured to determine push metric data of the resource to be recommended and the target object based on the first resource feature, the second resource feature, the first object feature, and the second object feature; A target recommended resource determination module, configured to push a target recommended resource to the target object; the target recommended resource is determined from the resources to be recommended based on the push metric data.

9. The device according to claim 8, characterized in that, The sample association features include multiple sample association sub-features, the second streaming feature extraction model includes association feature extraction models respectively corresponding to the multiple sample association sub-features, and the second feature extraction module includes: An association feature extraction unit, configured to input the resource information and the object information into each association feature extraction model for feature extraction, so as to obtain multiple association resource features of the resource to be recommended and multiple association object features of the target object; A first association feature splicing unit, configured to perform feature splicing processing on the multiple association resource features to obtain a second resource feature; A second association feature splicing unit, configured to perform feature splicing processing on the multiple association object features to obtain a second object feature.

10. The device according to claim 8, characterized in that The second streaming feature extraction model includes multiple layers of resource feature extraction layers and multiple layers of object feature extraction layers, the number of layers of the multiple layers of resource feature extraction layers is the same as the number of layers of the multiple layers of object feature extraction layers, and the second feature extraction module includes: A second resource feature extraction unit, configured to perform feature extraction based on the multiple layers of resource feature extraction layers and the resource information to obtain the second resource feature; A second object feature extraction unit, configured to perform feature extraction based on the multiple layers of object feature extraction layers and the object information to obtain the second object feature.

11. The device according to claim 8, characterized in that, The push metric determination module includes: A resource feature splicing unit, configured to perform feature splicing processing on the first resource feature and the second resource feature to obtain a target resource feature; An object feature splicing unit, configured to perform feature splicing processing on the first object feature and the second object feature to obtain a target object feature; A push metric data determination unit, configured to perform feature cross-processing based on the target resource feature and the target object feature to determine the push metric data.

12. The device according to claim 8, characterized in that, The apparatus further includes: A sample association feature determination module, configured to determine the sample association features based on a preset resource recommendation model; the preset resource recommendation model is obtained by training a model based on sample resource information of sample resources, sample object information of sample objects, and the association information between the sample objects and the sample resources; A training feature extraction module, configured to input the sample resource information and the sample object information into a model to be trained for feature extraction, so as to obtain training resource features corresponding to the sample resources and training object features corresponding to the sample objects; A feature association processing module, configured to perform feature association processing between the training resource features and the training object features to obtain training association features; A model training module, configured to perform model training on the model to be trained based on the sample association features and the training association features to obtain the second streaming feature extraction model.

13. The device according to claim 12, characterized in that, The sample association features include first sample association sub-features, the preset resource recommendation model includes a feature extraction layer, and the sample association feature determination module includes: A first sample association sub-feature determination module, configured to input the association information into the feature extraction layer for feature extraction to obtain first sample association sub-features; The second streaming feature extraction model includes a first association feature extraction model, and the model training module includes: A first loss information determination unit, configured to determine first loss information based on the first sample association sub-features and the training association features; A first model training unit, configured to perform model training on the model to be trained based on the first loss information to obtain the first association feature extraction model.

14. The device according to claim 12, characterized in that, The sample association features include second sample association sub-features, the preset resource recommendation model includes an attention learning layer, and the sample association feature determination module includes: A second sample association sub-feature determination module, configured to input the sample resource information, the sample object information, and the association information into the attention learning layer for attention learning to obtain second sample association sub-features; The second streaming feature extraction model includes a second association feature extraction model, and the model training module includes: A second loss information determination unit, configured to determine second loss information based on the second sample association sub-features and the training association features; A second model training unit, configured to perform model training on the model to be trained based on the second loss information to obtain the second association feature extraction model.

15. An electronic device, characterized in that The electronic device includes a processor and a memory, and at least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement a resource processing method according to any one of claims 1-7.

16. A computer-readable storage medium, characterized in that, The storage medium includes a processor and a memory, and at least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement a resource processing method according to any one of claims 1-7.

17. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the resource processing method according to any one of claims 1-7.

Images