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

Abstract

The application discloses a resource recommendation method and device, equipment, a storage medium and a program product. The method comprises the following steps: obtaining prediction data, the prediction data being used for predicting the interaction probability of a first account and a candidate resource, the prediction data comprising attribute feature data and scene feature data, the attribute feature data being used for indicating the attribute features of the first account and the candidate resource, and the scene feature data being used for indicating the scene features of the interaction between the first account and the candidate resource; determining a routing weight based on the scene feature data, the routing weight being used for indicating the influence degree of the attribute feature data on the interaction probability; predicting the interaction probability of the first account and the candidate resource based on the attribute feature data and the routing weight through a probability prediction model, the probability prediction model being a machine learning model that is pre-trained and used for predicting the occurrence probability of a specified event; and recommending the candidate resource to the first account based on the interaction probability, so that the prediction accuracy for different scenes is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a resource recommendation method, apparatus, device, storage medium, and program product. Background Technology

[0002] In the gaming platform ecosystem, recommending vouchers is a crucial step in boosting user activity and driving business conversions. Users redeem vouchers across a wide variety of scenarios, including various games' micro-community lifestyle benefit sections.

[0003] In related technologies, a unified model is used to process user data and voucher features. After the model outputs a user-voucher matching score, unified recommendations are made for the welfare zones of different games based on the ranking results.

[0004] However, because a uniform recommendation approach is difficult to adapt to the different needs of users in different scenarios, the prediction accuracy for differentiated scenarios is low, resulting in low recommendation efficiency and poor applicability of the model to different scenarios. Summary of the Invention

[0005] This application provides a resource recommendation method, apparatus, device, storage medium, and program product, which can improve the prediction accuracy and recommendation efficiency for differentiated scenarios. The technical solution is as follows.

[0006] On the one hand, a resource recommendation method is provided, the method comprising:

[0007] Acquire prediction data, which is used to predict the probability of interaction between the first account and the candidate resource. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resource. The scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resource.

[0008] The routing weight is determined based on the scene feature data, and the routing weight is used to indicate the degree of influence of the attribute feature data on the interaction probability.

[0009] The probability of interaction between the first account and the candidate resource is predicted by a probabilistic prediction model based on the attribute feature data and the routing weight. The probabilistic prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring.

[0010] If the interaction probability meets the recommendation requirements, the candidate resource is recommended to the first account based on the interaction probability.

[0011] On the other hand, a resource recommendation device is provided, the device comprising:

[0012] An acquisition module is used to acquire prediction data, which is used to predict the interaction probability between a first account and a candidate resource. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resource, and the scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resource.

[0013] The processing module is used to determine the routing weight based on the scene feature data, wherein the routing weight is used to indicate the degree of influence of the attribute feature data on the interaction probability;

[0014] The processing module is further configured to predict the interaction probability between the first account and the candidate resource based on the attribute feature data and the routing weight using a probability prediction model. The probability prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring.

[0015] The processing module is further configured to recommend the candidate resources to the first account based on the interaction probability if the interaction probability meets the recommendation requirements.

[0016] In some embodiments, the probabilistic prediction model includes at least two feature cross networks, each corresponding to a routing weight, and the routing weights are used to indicate the output weights of the feature cross networks.

[0017] The processing module is further configured to:

[0018] Information features are extracted from the attribute feature data by at least two feature cross networks in the probability prediction model, respectively, to obtain at least two feature cross results;

[0019] Based on the routing weights corresponding to the i-th feature cross result and the i-th feature cross network among the at least two feature cross results, the output result of the i-th network is determined, where i is a positive integer;

[0020] The interaction probability is predicted by the prediction network in the probability prediction model based on the network output results corresponding to the at least two feature cross results.

[0021] In some embodiments, the at least two feature cross networks include a low-order feature cross network and a high-order feature cross network;

[0022] The processing module is further configured to:

[0023] The low-order feature crossover network extracts low-order information features from the attribute feature data to obtain low-order feature crossover results.

[0024] The higher-order feature cross-network extracts higher-order information features from the attribute feature data to obtain higher-order feature cross-results.

[0025] Based on the low-order feature cross results and the low-order routing weights corresponding to the low-order feature cross network, the low-order network output results are determined. The low-order routing weights are determined based on the network parameters of the low-order feature cross network and the scene feature data.

[0026] Based on the higher-order feature cross results and the higher-order routing weights corresponding to the higher-order feature cross network, the output results of the higher-order network are determined. The higher-order routing weights are determined based on the network parameters of the higher-order feature cross network and the scene feature data.

[0027] In some embodiments, the processing module is further configured to:

[0028] The low-order routing weights are generated based on the scene feature data through the dynamic weight layer corresponding to the low-order feature cross network.

[0029] By concatenating the scene feature data and the attribute feature data, higher-order routing input data is obtained;

[0030] The higher-order routing weights are generated by the dynamic weight layer corresponding to the higher-order feature cross network based on the higher-order routing input data.

[0031] In some embodiments, the processing module is further configured to vectorize the attribute feature data to obtain an attribute feature vector, wherein the attribute feature vector includes at least one of dense feature vector, sparse feature vector and temporal feature vector, wherein the dense feature vector is used to indicate continuous numerical features in the attribute feature data, the sparse feature vector is used to indicate categorical features in the attribute feature data, and the temporal feature vector is used to indicate temporal features in the attribute feature data.

[0032] In some embodiments, the low-order feature cross network includes a low-order cross subnetwork and a high-order cross subnetwork;

[0033] The processing module is further configured to:

[0034] The low-order crossover subnetwork extracts information features based on the dense feature vector and the sparse feature vector to obtain the low-order crossover result. The low-order crossover subnetwork includes a first-order feature crossover network and a second-order feature crossover network. The first-order feature crossover network is used to quantify the degree of independent influence of the dense feature vector and the sparse feature vector on the interaction probability. The second-order feature crossover network is used to quantify the degree of joint influence of at least two dense feature vectors on the interaction probability.

[0035] By concatenating at least two of the attribute feature vectors, a higher-order cross-input feature is obtained;

[0036] The higher-order cross input features are processed layer by layer through the higher-order cross sub-network to obtain higher-order cross sub-results;

[0037] The low-order feature cross result is obtained by fusing the low-order cross result and the high-order cross result.

[0038] In some embodiments, the high-order feature cross-network includes a multi-expert cross-network;

[0039] The processing module is further configured to:

[0040] By concatenating the dense feature vector, the sparse feature vector, and the temporal feature vector, the input features of the expert network are obtained.

[0041] A general feature representation is extracted based on the input features of the expert network through a shared expert subnetwork in the multi-expert cross network;

[0042] Task feature representations are extracted based on the input features of the expert network through the task expert sub-network in the multi-expert cross network.

[0043] The higher-order feature cross result is obtained by fusing the general feature representation and the task feature representation.

[0044] In some embodiments, the processing module is further configured to:

[0045] By concatenating the network outputs corresponding to the at least two feature cross results, a fused feature is obtained;

[0046] The interaction probability is predicted by the prediction network based on the fusion features.

[0047] In some embodiments, the acquisition module is further configured to:

[0048] Account attribute data, resource attribute data, and context data are obtained as the attribute feature data. The account attribute data is used to indicate the account data of the first account, the resource attribute data is used to indicate the resource attributes of the candidate resources, and the context data is used to indicate the environmental state of the first account.

[0049] The program feature data and time feature data are obtained as the scene feature data. The program feature data is used to indicate the type of program that recommends the candidate resource to the first account, and the time feature data is used to indicate the time when the candidate resource is recommended to the first account.

[0050] In some embodiments, the acquisition module is further configured to acquire sample prediction data, which is used to predict the sample interaction probability of a sample account with a sample resource. The sample prediction data includes sample attribute data and sample scenario data, and the sample prediction data corresponds to a reference label.

[0051] The processing module is further configured to determine sample routing weights based on the sample scene data, wherein the sample routing weights are used to indicate the degree of influence of the sample attribute data on the sample interaction probability;

[0052] The processing module is further configured to predict the sample interaction probability of the sample account for the sample resource based on the sample attribute data and the sample routing weight using a candidate probability prediction model;

[0053] The processing module is further configured to train the candidate probability prediction model based on the difference between the sample interaction probability and the reference label until the training requirements are met, thereby obtaining the probability prediction model.

[0054] In some embodiments, the processing module is further configured to:

[0055] Obtain a first candidate resource whose interaction probability meets the recommendation requirements from the candidate resources;

[0056] The first candidate resource is recommended to the first account in descending order of the interaction probability, wherein the interaction probability of the first candidate resource is positively correlated with the recommendation display priority of the first candidate resource in the client where the first account is located, and the recommendation display priority is used to indicate the display order and visual prominence of the first candidate resource in the client where the first account is located.

[0057] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the resource recommendation method as described in any of the embodiments of this application above.

[0058] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, code set, or instruction set is stored therein, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the resource recommendation method as described in any of the embodiments of this application above.

[0059] On the other hand, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the resource recommendation methods described in the above embodiments.

[0060] The beneficial effects of the technical solutions provided in this application include at least the following:

[0061] By using the scene features of the interaction between the first account and candidate resources to participate in the prediction of interaction probability, the routing weight can be determined based on the scene feature data. This allows for the adjustment of the influence of attribute feature data on the interaction probability in different scenarios, improving the fit between the interaction probability predicted by the probability prediction model based on attribute feature data and the interaction scenario. This enhances the accuracy of the interaction probability prediction, resulting in higher recommendation efficiency for recommending candidate resources to the first account based on the interaction probability, and making it adaptable to differentiated recommendation scenarios. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a schematic diagram of a computer system provided in an exemplary embodiment of this application;

[0064] Figure 2 This is a flowchart of a resource recommendation method provided in an exemplary embodiment of this application;

[0065] Figure 3 This is a schematic diagram of the structure of a probability prediction model provided in an exemplary embodiment of this application;

[0066] Figure 4 This is a schematic diagram illustrating multi-scenario resource recommendation provided by an exemplary embodiment of this application;

[0067] Figure 5 This is a flowchart of an interactive probability prediction method provided in an exemplary embodiment of this application;

[0068] Figure 6 This is a schematic diagram of the structure of a high-order cross subnetwork provided in an exemplary embodiment of this application;

[0069] Figure 7 This is a schematic diagram of the structure of a multi-expert cross-network provided in an exemplary embodiment of this application;

[0070] Figure 8 This is a schematic diagram of the model structure of a probabilistic prediction model based on dynamic routing provided in an exemplary embodiment of this application;

[0071] Figure 9 This is a schematic diagram of a network structure provided in an exemplary embodiment of this application;

[0072] Figure 10 This is a schematic diagram of a network task provided in an exemplary embodiment of this application;

[0073] Figure 11 This is a flowchart of a training method for a probability prediction model provided in an exemplary embodiment of this application;

[0074] Figure 12 This is a structural block diagram of a resource recommendation apparatus provided in an exemplary embodiment of this application;

[0075] Figure 13 This is a structural block diagram of a terminal provided in an exemplary embodiment of this application. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0077] It should be understood that although the terms first, second, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, a first parameter may also be referred to as a second parameter without departing from the scope of this disclosure, and similarly, a second parameter may also be referred to as a first parameter. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0078] In the gaming platform ecosystem, coupon recommendations are a crucial step in enhancing user activity and monetization. Users redeem coupons across a wide variety of scenarios, including various games' micro-community lifestyle benefit zones. Related technologies utilize a unified model to process user data and coupon features. After the model outputs a user-coupon matching score, recommendations are uniformly delivered to different game benefit zones based on the ranking results. However, this uniform recommendation approach struggles to adapt to varying user needs across different scenarios, exhibiting low prediction accuracy for differentiated scenarios, resulting in low recommendation efficiency and poor model applicability across diverse situations.

[0079] The resource recommendation method provided in this application embodiment uses the scene features of the interaction between the first account and the candidate resources to participate in the prediction of the interaction probability. It can determine the routing weight based on the scene feature data, thereby adjusting the influence of attribute feature data on the interaction probability in different scenes according to the scene features. This improves the fit between the interaction probability predicted by the probability prediction model based on attribute feature data and the interaction scene, and improves the prediction accuracy of the interaction probability. As a result, the recommendation efficiency of recommending candidate resources to the first account based on the interaction probability is high, and it can adapt to differentiated recommendation scenarios.

[0080] First, the computer system described in this application will be introduced. Please refer to... Figure 1 The illustration shows a schematic diagram of a computer system provided in an exemplary embodiment of this application, which includes a terminal 10 and a server 20.

[0081] Terminal 10 has a first application installed. The first application has a resource recommendation function. The candidate resources recommended in the first application can be determined by server 20. Server 20 is the backend server of the first application and is used to determine the candidate resources recommended in the first application.

[0082] Optionally, the first application can be any application with resource recommendation function, such as a virtual game program, social application, online shopping software, multimedia playback platform, e-reading software, online teaching platform, virtual reality (VR) application, augmented reality (AR) application, mixed reality (MR) application, etc. This application embodiment does not limit this.

[0083] In some embodiments, server 20 is a backend server for at least two first applications, and server 20 is used to coordinate and determine recommended candidate resources in at least two first applications.

[0084] Taking the example of terminal 10 logging into at least two first applications through a first account, server 20 needs to recommend candidate resources to the first account in a targeted manner based on the account attributes of the first account and the program type of the first application. That is, the candidate resources recommended to the first account in at least two first applications may be the same or different.

[0085] Schematic, server 20 acquires prediction data, which is used to predict the interaction probability between the first account and candidate resources. The prediction data includes attribute feature data and scene feature data. The attribute feature data indicates the attribute features of the first account and candidate resources, and the scene feature data indicates the scene features of the interaction between the first account and candidate resources. Based on the scene feature data, a routing weight is determined, which indicates the degree of influence of the attribute feature data on the interaction probability. Based on the attribute feature data and the routing weight, a probability prediction model is used to predict the interaction probability between the first account and candidate resources. This probability prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring. If the interaction probability meets the recommendation requirements, candidate resources are recommended to the first account based on the interaction probability.

[0086] Among them, scene feature data can be used to recommend the program type of the first application as a candidate resource to the first account.

[0087] The aforementioned terminal is optional and can be a desktop computer, laptop computer, mobile phone, tablet computer, e-book reader, Moving Picture Experts Group Audio Layer III (MP3) player, Moving Picture Experts Group Audio Layer IV (MP4) player, smart TV, smart vehicle, and other types of terminal devices. This application embodiment does not limit the specific terminal device to these types.

[0088] It is worth noting that the aforementioned servers can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers that provide basic cloud computing services such as cloud services, cloud security, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0089] Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to achieve data computing, storage, processing, and sharing.

[0090] In some embodiments, the server described above can also be implemented as a node in a blockchain system.

[0091] It should be noted that all 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 involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant regions. For example, the operational data and account information involved in this application were obtained with full authorization.

[0092] To further clarify, this application may display a prompt interface, pop-up window, or output voice prompts before and during the collection of user-related data (e.g., account information, historical operation data, and real-time operation data involved in this application). These prompt interfaces, pop-ups, or voice prompts are used to inform the user that their relevant data is being collected. This ensures that the application only begins the steps for collecting user-related data after receiving confirmation from the user regarding the prompt interface or pop-up window; otherwise (i.e., without receiving confirmation from the user), the steps for collecting user-related data end, meaning no user-related data is collected. In other words, all user data collected in this application is collected with the user's consent and authorization, and the collection, use, and processing of relevant user data must comply with the relevant laws, regulations, and standards of the relevant regions.

[0093] This is illustrative; please refer to it. Figure 2 This document illustrates a flowchart of a resource recommendation method provided in an exemplary embodiment of this application. This method can be executed by a terminal, a server, or both simultaneously. This embodiment uses server execution as an example for illustration. Figure 2 As shown, the method includes the following steps:

[0094] Step 210: Obtain the prediction data.

[0095] The prediction data is used to predict the probability of interaction between the first account and the candidate resources. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resources, and the scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resources.

[0096] Taking the recommendation of candidate resources to the first account in the first application as an example, the candidate resources can be any resources authorized and recommended by the first application, such as virtual game resources, virtual vouchers, multimedia playback permissions, e-reading permissions, social interaction permissions, etc. The resources authorized and recommended by the first application include, but are not limited to, the candidate resources provided in the first application, or at least one of the candidate resources recommended by the second application authorized and accessed by the first application.

[0097] The second application is an application authorized by the first application, which can access the first application and recommend candidate resources within the first application.

[0098] The interaction between the first account and the candidate resources includes, but is not limited to, any interaction method such as the first account viewing the candidate resources, the first account clicking on the candidate resources, the first account claiming the candidate resources, and the first account using the candidate resources. This application embodiment does not limit this.

[0099] Taking virtual vouchers as an example of candidate resources, the interaction probability can be the probability of the first account clicking on the virtual voucher, the probability of the first account receiving the virtual voucher, or the probability of the first account redeeming the virtual voucher, etc.

[0100] The scene features indicated by the scene feature data include, but are not limited to, at least one of the following: the program type of the first application that recommends candidate resources to the first account, or the recommendation time of the candidate resources to the first account.

[0101] In some embodiments, obtaining prediction data includes obtaining attribute feature data and scene feature data. Account attribute data, resource attribute data, and context data are obtained as attribute feature data. Account attribute data is used to indicate the account data of the first account, resource attribute data is used to indicate the resource attributes of the candidate resources, and context data is used to indicate the environmental state of the first account. Program feature data and time feature data are obtained as scene feature data. Program feature data is used to indicate the program type of the candidate resources recommended to the first account, and time feature data is used to indicate the recommendation time of the candidate resources recommended to the first account.

[0102] To acquire attribute feature data, account attribute data, resource attribute data, and context data can be collected in real time from the business database.

[0103] For illustrative purposes only, account attribute data includes, but is not limited to, user age, gender, account level, and historical behavior sequence; resource attribute data includes, but is not limited to, coupon identifier, brand category, discount level, and click-through rate; contextual data includes, but is not limited to, the device type, geographical location, and network environment corresponding to the first account.

[0104] Scene feature data includes program features and time features. Program features include, but are not limited to, the program category of the first application and the channel identifier for recommending candidate resources to the first account. Time features include, but are not limited to, the date, holiday type, and time period for recommending candidate resources to the first account.

[0105] In some embodiments, feature preprocessing is performed: normalizing continuous numerical features (such as click rate), performing Embedding mapping on categorical features (such as game ID), and encoding behavior sequence features (such as recent click records) through the Attention mechanism.

[0106] Step 220: Determine routing weights based on scene feature data.

[0107] The routing weight is used to indicate the degree of influence of attribute feature data on the interaction probability.

[0108] In recommendation scenarios, there are significant differences in users' preferences for resources across different scenarios. Therefore, it is necessary to consider the scenario where resources are recommended to users, so as to achieve differentiated recommendation.

[0109] For illustration, taking differences in game scenarios as an example, players of virtual game A may prefer dining coupons, while players of virtual game B may pay more attention to discounts on digital products; taking differences in time scenarios as an example, users prefer fast food coupons at noon on workdays, while they may prefer cinema discounts on weekend evenings.

[0110] In related technologies, a single model is used to process all scenarios, which limits the recommendation effect. The method provided by the embodiments of the present application enables the model to automatically adjust the importance of different features according to scenario features through a dynamic routing weight mechanism, improving recommendation accuracy.

[0111] Specifically, the routing weights are determined based on the scenario feature data, so that the routing weights can determine the degree of influence of the attribute feature data on the interaction probability based on the scenario features, thereby realizing scenario awareness, and differentiating the recommendation results obtained by the same user in different scenarios, for example, recommending coffee coupons on workdays and restaurant set meals on weekends; on the other hand, feature-level dynamic adjustment can also be realized, for example, during game promotions, the weight of the "discount intensity" feature is automatically increased; and by generating weights, only a small number of parameters need to be added to the probability prediction model, without retraining a new model for each different scenario, which improves the recommendation efficiency.

[0112] In some embodiments, the routing weight can be obtained by adding a gating mechanism in the probability prediction model. When training the probability prediction model, by assigning reference labels to the training data, the model can learn the degree of influence of different features in the training data on the prediction results under different scenarios while learning the probability prediction capability, so that the trained probability prediction model can automatically generate routing weights based on the scenario feature data through the gating mechanism, realizing an automatic dynamic routing mechanism.

[0113] Step 230: Predict the interaction probability between the first account and the candidate resource through the probability prediction model based on the attribute feature data and the routing weight.

[0114] A probabilistic prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring.

[0115] In some embodiments, when a resource recommendation scenario is triggered by the client, the client can automatically call the feature service, provide attribute feature data such as request and product features, obtain the corresponding feature vector based on the attribute feature data, call the model service, which is used to provide a pre-trained probability prediction model, perform dynamic routing calculation and multi-connection feature cross based on the feature vector, obtain the interaction probability prediction result, determine the recommendation result based on the interaction probability prediction result and return it to the client.

[0116] This is illustrative; please refer to it. Figure 3 , Figure 3 This is a schematic diagram of the structure of a probability prediction model provided in an exemplary embodiment of this application, as shown below. Figure 3 As shown, the probability prediction model 300 includes a feature cross-layer 310 and a dynamic routing layer 320. The feature cross-layer 310 is used to extract cross-features that affect the interaction probability based on attribute feature data. The dynamic routing layer 320 is used to automatically generate routing weights based on scene feature data, so that the interaction probability predicted based on cross-features and routing weights conforms to the interaction scene.

[0117] In some embodiments, in order to accurately predict interaction probabilities based on attribute feature data, the probability prediction model includes at least two feature cross networks, each corresponding to a routing weight, which is used to indicate the output weight of the feature cross network.

[0118] Indicatively, information features are extracted from attribute feature data by at least two feature cross networks in the probabilistic prediction model to obtain at least two feature cross results; the output result of the i-th network is determined based on the routing weights corresponding to the i-th feature cross result and the i-th feature cross network, where i is a positive integer; the interaction probability is predicted by the prediction network in the probabilistic prediction model based on the network output results corresponding to the at least two feature cross results.

[0119] In some embodiments, at least two feature cross networks include a low-order feature cross network and a high-order feature cross network.

[0120] In a schematic manner, low-order information features are extracted from attribute feature data through a low-order feature cross-network to obtain low-order feature cross-results; high-order information features are extracted from attribute feature data through a high-order feature cross-network to obtain high-order feature cross-results; based on the low-order feature cross-results and the corresponding low-order routing weights of the low-order feature cross-network, the output result of the low-order network is determined; based on the high-order feature cross-results and the corresponding high-order routing weights of the high-order feature cross-network, the output result of the high-order network is determined; the output results of the low-order network and the high-order network are concatenated to obtain fused features, and the interaction probability is predicted by a prediction network based on the fused features.

[0121] Low-order feature cross results are representation vectors generated by explicitly modeling the basic interaction relationships between features (such as first-order linear terms and second-order combination terms).

[0122] The low-order feature cross results are generated based on dense and sparse features in the attribute feature data, including first-order linear terms, second-order combination terms, and higher-order cross terms. The first-order linear terms are used to indicate the degree of independent influence of a single feature on the interaction probability, the second-order combination terms are used to indicate the pairwise latent vector product of sparse features, and the higher-order cross terms are used to indicate explicit element-wise crosses while retaining the original features.

[0123] The high-order feature cross result is a deep representation vector generated by implicitly learning the complex nonlinear relationships (third-order and above combinations) between features.

[0124] The high-order feature cross result is generated by splicing together all dense features, sparse features and temporal features in the attribute feature data.

[0125] Illustratively, low-order feature cross networks can include Factorization Machines (FM) and Deep & Cross Networks (DCN), such as Deep & Cross Networks version 2 (DCNv2), while high-order cross networks can include Shared and Task-specific Embeddings (STEM).

[0126] FM can be used to implement low-order explicit crosses of sparse features, such as handling the impact of basic attribute features like account identifiers and resource identifiers on interaction probabilities; DCNv2 is used to implement high-order explicit crosses of dense or sparse features, modeling complex feature combinations, such as time series × category; STEM is used to implement multi-task dynamic routing and high-order implicit crosses, adapting to multi-task objectives such as click-through rate / redeem rate.

[0127] Schematic, the FM network includes first-order and second-order terms. The first-order term calculates the linear effect, such as user age multiplied by discount level; the second-order term calculates the latent vector inner product of user identifier and voucher identifier. The DCNv2 network multiplies the original features element-wise with the output of the cross layer, generating a high-order feature combination through three layers of cross processing. The STEM network extracts general representations of user interests through shared experts, predicts short-term behavioral characteristics of experts based on click-through rates among task experts, predicts long-term consumption preferences based on redemption rates among task experts, and dynamically mixes expert outputs according to the current scenario through a gating network. After feature cross processing, the prediction network concatenates the low-order / high-order cross results, and the interaction probability is output through a two-layer fully connected network.

[0128] It is worth noting that the structure and data processing of the above probability prediction model are merely illustrative examples, and the embodiments of this application do not limit them.

[0129] Step 240: If the interaction probability meets the recommendation requirements, recommend candidate resources to the first account based on the interaction probability.

[0130] Optionally, candidate vouchers can be sorted by predicted probability, or candidate resources with interaction probabilities sorted from high to low within a preset ranking range can be pushed to the user interface.

[0131] In some embodiments, a first candidate resource whose interaction probability meets the recommendation requirements is obtained from the candidate resources; the first candidate resource is recommended to the first account in descending order of interaction probability, wherein the interaction probability corresponding to the first candidate resource is positively correlated with the recommendation display priority of the first candidate resource in the client where the first account is located, and the recommendation display priority is used to indicate the display order and visual prominence of the first candidate resource in the client where the first account is located.

[0132] For illustrative purposes, the recommendation requirement includes an interaction probability of 70%. The first candidate resource with an interaction probability of 70% is selected from the candidate resources and recommended to the first account. When recommending the first candidate resource to the first account, the server can send the first candidate resource to the client of the first account. The first candidate resource is associated with a recommendation display priority, which is positively correlated with the interaction probability. The client of the first account can prioritize rendering the first candidate resource with the higher recommendation display priority during the interface loading process, and display the first candidate resource with higher visual prominence according to the recommendation display priority. For example, according to the interface rendering template of the resource recommendation interface in the first application, the first candidate resource with the higher recommendation display priority is prioritized and displayed in a prominent position.

[0133] This is illustrative; please refer to it. Figure 4 , Figure 4 This is a schematic diagram illustrating multi-scenario resource recommendation provided in an exemplary embodiment of this application, such as... Figure 4 As shown, a server is jointly recommended for both game program A and game program B. During resource recommendation for game program A, the game interface 410 of game program A is displayed. In response to a trigger operation on the first control 411, a first resource recommendation interface 420 is displayed in game program A. The first resource recommendation interface 420 includes candidate resources 421 recommended to the first account. The candidate resources 421 are recommended based on the interaction probability determined by the attribute feature data and scene feature data of the first account's interaction with candidate resources in game program A. During resource recommendation for game program B, in response to a trigger operation on the second control 431 in the game interface 430 of game program B, a second resource recommendation interface 440 is displayed in game program B. The second resource recommendation interface 440 includes candidate resources 441 recommended to the first account. The candidate resources 441 are recommended based on the interaction probability determined by the attribute feature data and scene feature data of the first account's interaction with candidate resources in game program B.

[0134] In summary, the method provided in this application, by using the scene features of the interaction between the first account and the candidate resources to participate in the prediction of the interaction probability, can determine the routing weight based on the scene feature data, thereby adjusting the influence of attribute feature data on the interaction probability in different scenarios according to the scene features, improving the fit between the interaction probability predicted by the probability prediction model based on attribute feature data and the interaction scenario, improving the prediction accuracy of the interaction probability, and thus the recommendation efficiency of recommending candidate resources to the first account based on the interaction probability is high, which can adapt to differentiated recommendation scenarios.

[0135] In some embodiments, the probabilistic prediction model includes at least two feature cross-networks, each corresponding to a routing weight, which indicates the output weight of the feature cross-network. Please refer to [reference needed]. Figure 5 , Figure 5 This is a flowchart of an interactive probability prediction method provided in an exemplary embodiment of this application. This method can be executed by a terminal, a server, or both simultaneously. This embodiment illustrates the method as being executed by a server. Figure 5 As shown, step 230 above includes the following steps:

[0136] Step 231: Extract information features from the attribute feature data through at least two feature cross networks in the probability prediction model to obtain at least two feature cross results.

[0137] At least two feature cross networks are used to perform feature cross processing on attribute feature data from different dimensions.

[0138] In some embodiments, the attribute feature data needs to be characterized first. This process includes feature vectorization of the attribute feature data to obtain attribute feature vectors.

[0139] The attribute feature vector includes at least one of dense feature vector, sparse feature vector and temporal feature vector. Dense feature vector is used to indicate continuous numerical features in attribute feature data, sparse feature vector is used to indicate categorical features in attribute feature data, and temporal feature vector is used to indicate temporal features in attribute feature data.

[0140] Indicatively, dense feature vectors can be used to indicate the click-through rate, redemption rate, user age, and number of days logged into the game for a certain type of voucher on a given day; sparse feature vectors can be used to indicate user gender, location, and the type of voucher brand; and time-series feature vectors can be used to indicate the historical sequence of vouchers received by a user and the historical sequence of brands exposed to the user.

[0141] By vectorizing attribute feature data into dense, sparse, and temporal feature vectors, the characteristics of attribute feature data in continuous numerical, categorical, and temporal dimensions can be fully reflected. This provides a foundation for implementing at least two feature cross-processing methods, enabling at least two feature cross-processing networks to fully learn the complex relationships between attribute feature data based on three-dimensional feature vectors, thereby improving the accuracy of predicting interaction probabilities based on attribute feature data. Multimodal feature support is achieved by being compatible with numerical, categorical, and temporal data, and the problem of feature heterogeneity is solved by vectorization to achieve unified data representation.

[0142] This application embodiment takes at least two feature cross-networks, including a low-order feature cross-network and a high-order feature cross-network, as an example. The low-order feature cross-network is used to explicitly model the basic interaction relationship between attribute features, and the high-order feature cross-network is used to implicitly capture the deep nonlinear interaction pattern between attribute features.

[0143] To illustrate, low-order information features are extracted from attribute feature data through a low-order feature cross network to obtain low-order feature cross results; high-order information features are extracted from attribute feature data through a high-order feature cross network to obtain high-order feature cross results.

[0144] In this application embodiment, a low-order feature cross network including a low-order cross sub-network and a high-order cross sub-network, and a high-order feature cross network including a multi-expert cross network are used as examples. The low-order cross sub-network is used to capture explicit low-order interaction relationships between attribute features, such as first-order linear relationships and second-order combination relationships, mainly to solve the basic correlation problem under sparse features. The high-order cross sub-network is used to model implicit high-order nonlinear relationships (third-order and above complex combinations) between features and to mine deep feature patterns.

[0145] Optionally, the low-order cross subnetwork includes a first-order feature network and a second-order feature network. The first-order feature network is used to quantify the independent impact of a single feature, such as the linear contribution of user age to click-through rate. The second-order feature network is used to model pairwise feature interactions through latent vector inner product, such as the combined effect of user identifier × product identifier.

[0146] In some embodiments, information features are extracted based on dense and sparse feature vectors through a low-order cross sub-network to obtain a low-order cross sub-result. The low-order cross sub-network includes a first-order feature cross network and a second-order feature cross network. The first-order feature cross network is used to quantify the independent influence of dense and sparse feature vectors on the interaction probability, and the second-order feature cross network is used to quantify the joint influence of at least two dense feature vectors on the interaction probability. At least two of the attribute feature vectors are concatenated to obtain high-order cross input features. The high-order cross input features are then processed layer by layer through the high-order cross sub-network to obtain a high-order cross sub-result. Finally, the low-order cross sub-result and the high-order cross sub-result are fused to obtain a low-order feature cross result.

[0147] As an illustration, low-order cross-subnetworks include FM networks. The core idea of ​​FM networks is to capture the interaction relationships between features through factorization. Please refer to Equation 1 below for the model principle.

[0148]

[0149] Where, x i It is the input feature, v i It is the i-th input x i The embedding vector, v j It is the j-th input x j The embedding vector, w i w0 and w0 are the weights and biases, respectively. This formula uses two distinct features x... i With x j The inner product of the corresponding Embedding vectors is used as the second-order feature x. i x j The calculated weights explicitly capture the interaction relationships between features.

[0150] Higher-order cross subnetworks include DCNv2 networks; please refer to [link / reference]. Figure 6 , Figure 6 This is a schematic diagram of the structure of a high-order cross subnetwork provided in an exemplary embodiment of this application, as shown below. Figure 6As shown, the high-order cross sub-network 600 includes a stacked structure 610 and a parallel structure 620. In the stacked structure 610, the cross layer and the deep layer are stacked in sequence to form a deep network. In the parallel structure 620, the cross layer and the deep layer process the input features in parallel and then merge the results.

[0151] The input layer of DCNv2 is used to input raw features, which typically include numerical and categorical features. Categorical features are usually transformed into dense vectors through an embedding layer.

[0152] DCNv2's Cross Layer is used to explicitly capture higher-order interactions between features. The processing logic of the Cross Layer is shown in Equation 2 below.

[0153] x l+1 =x0⊙(W l x l +b l )+x l Formula 2,

[0154] Where x0 is the input feature, x l It is the output of the l-th layer, W l and b l These are the weights and biases, respectively. This formula modifies the input feature x0 by comparing it with the output x0 of the current layer. l Perform cross-products to explicitly capture the interactions between features.

[0155] DCNv2's deep layers consist of multiple fully connected layers for automatically learning feature representations. By introducing non-linear activation functions and a multi-layered structure, deep layers are able to capture complex patterns in the data.

[0156] Taking a high-order feature cross network, including a multi-expert cross network (STEM), as an example, dense feature vectors, sparse feature vectors, and temporal feature vectors are concatenated to obtain the expert network input features; a general feature representation is extracted based on the expert network input features through a shared expert sub-network in the multi-expert cross network; a task feature representation is extracted based on the expert network input features through a task expert sub-network in the multi-expert cross network; and the general feature representation and the task feature representation are fused to obtain the high-order feature cross result.

[0157] Higher-order feature cross-networks include STEM networks; please refer to [link / reference]. Figure 7 , Figure 7 This is a schematic diagram of the structure of a multi-expert cross-network provided in an exemplary embodiment of this application, such as... Figure 7As shown, the high-order feature cross-network 700 includes an embedding layer 710, an expert layer 720, a gating layer 730, and an output layer 740.

[0158] The embedding level of STEM is used to input the original features, which typically include numerical and categorical features. These features are then transformed into dense vectors through the embedding level. Here, a different set of dense vectors is generated for different tasks and shared experts.

[0159] The STEM Expert Level is used to capture the interaction relationships between features and includes multiple fully connected layers. Experts are further divided into task-specific experts and shared experts. Each expert module consists of multiple fully connected layers and can capture different data features respectively.

[0160] The gate level in STEM is determined by the number of tasks and is responsible for aggregating the outputs of multiple different experts. Its output is a weighted sum of the outputs of multiple different experts, and the weights are often obtained by passing the input embedding dense vector through a gated network and a softmax activation function.

[0161] The output layer (Tower Level) of STEM is determined by the number of tasks and includes multiple fully connected layers, each used to output the model's prediction y for task A. A And the prediction of task B y B .

[0162] By employing latent vector decomposition of low-order cross subnetworks (FM) to reduce the number of parameters and achieve sparse feature optimization, and by supporting cross-networks of arbitrary order (DCNv2) to achieve high-order scalability.

[0163] Multi-expert networks enable multi-task learning, with shared expert networks extracting general features and task expert networks optimizing specific objectives. They can also address the negative transfer problem. Through gating mechanisms, expert weights are adaptively allocated, enabling dynamic feature selection in multi-expert networks.

[0164] Step 232: Based on the routing weights corresponding to the i-th feature cross-network among at least two feature cross-network results, determine the output result of the i-th network, where i is a positive integer.

[0165] In some embodiments, the low-order network output is determined based on the low-order feature cross results and the low-order routing weights corresponding to the low-order feature cross network; the high-order network output is determined based on the high-order feature cross results and the high-order routing weights corresponding to the high-order feature cross network.

[0166] Among them, the low-order routing weights are determined based on the network parameters and scene feature data of the low-order feature cross network; the high-order routing weights are determined based on the network parameters and scene feature data of the high-order feature cross network.

[0167] In some embodiments, low-order routing weights are determined based on network parameters and scene feature data of low-order feature cross-networks; high-order routing weights are determined based on network parameters, scene feature data, and attribute feature data of high-order feature cross-networks.

[0168] Step 233: The interaction probability is predicted by the prediction network in the probability prediction model based on the network output results corresponding to at least two feature cross results.

[0169] In some embodiments, the network outputs corresponding to at least two feature cross-results are concatenated to obtain fused features, and the interaction probability is predicted by a prediction network based on the fused features.

[0170] This is illustrative; please refer to it. Figure 8 , Figure 8 This is a schematic diagram of the model structure of a probabilistic prediction model based on dynamic routing provided in an exemplary embodiment of this application, as shown below. Figure 8 As shown, the probability prediction model 800 includes a low-order feature cross-network 810, a high-order feature cross-network 820, and a prediction network 830.

[0171] The resource recommendation method provided in this application, in determining the interaction probability between a first account and candidate resources, employs, as follows: Figure 8 The probabilistic prediction model 800 shown executes steps 1 to 6, including step 1, information collection; step 2, feature vectorization; step 3, vector concatenation; step 4, information extraction; step 5, network task; and step 6, task prediction.

[0172] Regarding the information collection process, collect data that may be relevant to conversion rate prediction.

[0173] It is illustrative to collect basic information about the brands associated with the recommended vouchers, basic information about the vouchers themselves, statistical information such as the vouchers' recent click-through rate and conversion rate, as well as basic information about the users and their behavioral sequences.

[0174] For the feature vectorization process, feature data is mapped to a high-dimensional space.

[0175] Intuitively, the information collected in the first step is divided into three categories: dense features, sparse features, and sequence features. Dense features are continuous numerical features, such as the click-through rate and redemption rate of a certain type of coupon on a given day, the user's age, and the number of days they logged into the game. Sparse features are categorical features, such as user gender, location, and the brand category of the coupon. Sequence features are time-related data, such as the historical sequence of coupons a user has received or the historical sequence of brands a user has been exposed to. These features are vectorized and transformed into input tensors for the model. Both sparse and dense features use lookup embedding, representing each feature with a 16-dimensional feature tensor. Since the numerical values ​​of dense features have specific meanings, the vectorized dense features are multiplied by their numerical values ​​to obtain the final result. Sequence features, due to their temporal relationship, utilize an attention mechanism to extract deep features. In the attention mechanism, query is brand_id, and key is the sequence of brand_ids for user exposure, claiming, consumption, and subscription. The final sequence tensor representation can be calculated using the formula: Attention(query,key,value)=softmax(MLP(query,key))*value.

[0176] The sequence information contains users' historical behavior regarding in-game voucher services, reflecting their overall interests. Therefore, the feature module uses user behavior sequences to represent shared information. When using the Attention mechanism for information extraction, the query will differ depending on the game the user is playing. The query information includes not only the brand information to be recommended but also information about the current game. This allows for the full extraction of shared interest features from the user sequence and the differences between different games.

[0177] For the vector concatenation process, vectors are concatenated into a matrix vector according to a unified dimension.

[0178] To illustrate, based on actual business needs, vector concatenation mainly consists of three parts: using dense feature vectors as input to the SENet & DCNv2 network; concatenating sparse feature vectors and dense feature vectors to form a new vector as input to the SENet & FM network; and concatenating sparse feature, dense feature, and sequence vectors to form a new vector as input to the STEM network.

[0179] For the information extraction process, low-order feature cross-networks and high-order feature cross-networks are used to extract information respectively.

[0180] Indicatively, the system fully extracts information from the input features, mainly divided into three parts: the first-order feature crossover (FM) module, the second-order feature crossover (DCNv2) module, and the high-order feature crossover (Deep) module. At the same time, the SENet module is used for automatic feature selection, and two scene information gating modules are used to perceive the differences in scene distribution.

[0181] This paper introduces the scene information gating network for low-order cross-modules.

[0182] To enhance the importance of scene features, the output of the first-order feature cross of the FM module and the output of the second-order feature cross of DCNv2 are concatenated and then input into a scene feature layer for enhancement. Please refer to Formulas 3 and 4 below for details.

[0183] α s =stop_gradient[σ(W s X s +b s )×2] Formula 3,

[0184] O s =Concat(X) FM ,X DCMv2 )*α s Formula 4,

[0185] Among them, X s It is a scene feature, W s and b s These are learnable parameters, σ() is the Sigmoid activation function, stop_gradient stops backpropagation of gradients, and α... s X represents the scene feature weights, that is, the low-order routing weights determined based on scene feature data. FM It is a first-order feature cross-output, X DCMv2 It is a second-order feature cross-output; Concat is a tensor concatenation operation, O s The final output of the scene information gating network representing the low-order cross-modules is represented by the symbols × and *, respectively, for scalar multiplication and element-wise multiplication.

[0186] This paper introduces scene information gating networks for high-order cross-models.

[0187] The high-order feature cross-network adopts the STEM model, which is a multi-layer, multi-expert network. For the input X of the l-th layer... l The output O of task T T It will be calculated jointly by n experts. If we denote Expert as... iIf it is the i-th expert task network, then O T The specific calculation formulas are as follows: Formulas 5 to 7.

[0188] X Ei =Expert i (X l Formula 5, i = 1, ..., n

[0189] β T =Φ(W T ·Concat(X l ,X S )+b T ) Formula 6,

[0190]

[0191] Where X S It is a scene feature, W T and b T These are learnable parameters; Concat is a tensor concatenation operation; and Φ(x) is the Softmax activation function. Represents the task gating weights that define the scene characteristics. This represents the output feature of the i-th expert network.

[0192] The probabilistic prediction model described above also utilizes a squeeze-and-excitation network (SENet). For an illustrative example, please refer to [link / reference]. Figure 9 , Figure 9 This is a schematic diagram of a network structure provided in an exemplary embodiment of this application, such as... Figure 9 As shown, the network structure 900 includes the SENet network, corresponding to the Squeeze stage 910, the Excitation stage 920, and the Scale stage 930.

[0193] SENet primarily performs a squeeze operation on the input to generate weights, which are then combined with the original input element-wise. For an input SENet tensor of shape (C, H, W), max-pooling yields a weight tensor of dimension (1, 1, C). This weight tensor is then passed through two dense layers with ReLU and Sigmoid activation functions, and used as weights along dimension C for element-wise multiplication with the original input tensor. Essentially, SENet automatically filters the input tensor along dimension C.

[0194] By generating low-order routing weights using scene feature data for low-order feature cross networks and generating high-order routing weights using concatenated features of scene feature data and attribute feature data for high-order feature cross networks, the paths generated for different feature cross networks can be matched with the feature cross order in the feature cross network, thereby improving the accuracy of dynamic routing.

[0195] For network tasks, please refer to Figure 10 , Figure 10 This is a schematic diagram of a network task provided in an exemplary embodiment of this application, such as... Figure 10 As shown, input data 1010 is received through the input layer, which includes a one-dimensional or multi-dimensional tensor; the input data 1010 is processed by a fully connected layer and an activation function to obtain mapped data 1020; the output layer outputs based on the mapped data 1020 to obtain output data 1030, which is one-dimensional data.

[0196] Specifically, after extracting information from the model input through different modules, the resulting tensors are concatenated and then input into the network task module. The network task module contains multiple fully connected layers, which essentially reduce the dimensionality of the extracted tensors layer by layer, ultimately obtaining a 1-dimensional output, denoted as logit, which serves as the output of the network task.

[0197] By concatenating the outputs of multiple networks and then using them for prediction, it is possible to integrate low-order and high-order cross-results, thereby improving the model's expressive power.

[0198] After obtaining the logit output, the Sigmoid activation function transforms it into a value between 0 and 1, which serves as the model's prediction of the coupon's click-through rate. From this, the binary cross-entropy loss function can be calculated; the specific calculation method is shown in Formula 8 below.

[0199]

[0200] Where y represents the real label. To predict the interaction probabilities, the model parameters are continuously updated during training through gradient backpropagation.

[0201] In summary, the method provided in this application, through the collaborative work of at least two feature cross-networks, fully captures the interaction between low-order and high-order features. Furthermore, by employing an automatic routing mechanism based on scene feature data, it dynamically selects information transmission paths within the neural network, better capturing complex patterns and relationships in the data. This allows the system, based on at least two feature cross-networks, to select information transmission paths more suited to the interaction scenario according to dynamic weight allocation, improving the adaptability of prediction results to the interaction scenario and thus enhancing prediction accuracy. On the other hand, dynamic weight allocation avoids negative transfer and improves the model's generalization ability.

[0202] The method provided in this application, when generating low-order feature cross results and high-order feature cross results through low-order feature cross networks and high-order feature cross networks respectively, adjusts the network output results by using low-order routing weights and high-order routing weights corresponding to the low-order feature cross networks and high-order feature cross networks respectively, and participates in interaction probability prediction. This enables the probability prediction model to focus more on the feature cross networks that perform better in the corresponding scenarios when predicting interaction probabilities, thereby improving the adaptability to multi-scenario prediction.

[0203] In some embodiments, the resource recommendation method provided in this application further includes a training process for a probabilistic prediction model. Please refer to... Figure 11 This document illustrates a flowchart of a training method for a probabilistic prediction model provided in an exemplary embodiment of this application. This method can be executed by a terminal, a server, or both simultaneously. This embodiment uses server execution as an example for illustration. Figure 11 As shown, the method includes the following steps:

[0204] Step 1110: Obtain sample prediction data.

[0205] The sample prediction data is used to predict the probability of sample account interaction with sample resources. The sample prediction data includes sample attribute data and sample scenario data, and the sample prediction data has corresponding reference labels.

[0206] Taking the prediction of user interaction probability with resources in at least two first applications as an example, sample prediction data can be obtained from the business logs of at least two first applications. The sample prediction data may include historical business data and real-time business data of at least two first applications.

[0207] For illustrative purposes, the sample attribute data includes dense features (such as click-through rate, user level, etc.), sparse features (such as user identifier, product identifier, etc.), and time-series features (including behavior sequences). The sample scenario data includes the recommendation channel identifier and recommendation time period of the recommended resources. The reference label can be used to indicate whether the user clicks on the recommended resource, and the value is 0 or 1.

[0208] Step 1120: Determine the sample routing weights based on the sample scenario data.

[0209] Sample routing weights are used to indicate the degree of influence of sample attribute data on sample interaction probabilities.

[0210] In some embodiments, the candidate probability prediction model includes a low-order feature cross network and a high-order feature cross network, and low-order sample routing weights and high-order sample routing weights need to be generated for the low-order feature cross network and the high-order feature cross network, respectively.

[0211] For the generation process of low-order sample routing weights, we can generate low-order sample routing weights by taking only scene features as input for generative networks including single-layer fully connected and Sigmoid activation. During backpropagation, we can freeze the gradient of the weight generation layer to avoid affecting the parameters of the feature-crossing network.

[0212] Specifically, please refer to formulas 3-4 above, which will not be repeated here.

[0213] For the generation process of high-order sample routing weights, a generative network consisting of two fully connected layers and Softmax activation can be used. The concatenated features of scene feature data and attribute feature data are input to generate high-order sample routing weights. During gradient propagation, only the high-order sample weights are allowed to influence the expert network selection, and they are not backpropagated to the feature encoding layer.

[0214] Specifically, please refer to formulas 5-7 above, which will not be repeated here.

[0215] Step 1130: Predict the sample interaction probability of a sample account for a sample resource based on sample attribute data and sample routing weights using a candidate probability prediction model.

[0216] The prediction process includes a feature cross-processing process and a probability prediction process. The feature cross-processing process includes low-order cross-processing and high-order cross-processing. Illustratively, taking candidate probability prediction models including the FM network, DCNv2 network, and STEM network to be trained as an example, the low-order cross-processing process uses the FM network to process first- and second-order feature cross-processing, and DCNv2 to process third-order feature cross-processing. The network outputs of the FM network and DCNv2 network are then weighted and fused according to the sample routing weights. The high-order cross-processing process uses the STEM network to perform feature cross-processing based on the task network and expert network respectively. The network outputs of the task network and expert network are dynamically weighted according to the sample routing weights.

[0217] Step 1140: Train the candidate probability prediction model based on the difference between the sample interaction probability and the reference label until it meets the training requirements, and obtain the probability prediction model.

[0218] During training, using `stop_gradient` stops gradient backpropagation, preventing routing weights from affecting the adjustment of model parameters in the feature cross model.

[0219] For illustration, you can refer to Formula 8 above to calculate the predicted loss value based on the difference between the sample interaction probability and the reference label, and train the candidate probability prediction model based on the predicted loss value. This will not be elaborated further here.

[0220] Optionally, training requirements include, but are not limited to, at least one of the following: convergence of the predicted loss value, or reaching a preset threshold number of training iterations. This application embodiment does not limit these requirements.

[0221] In summary, the method provided in this application embodiment realizes an end-to-end training optimization model training scheme for routing weights and model parameters, which can ensure the scene perception capability of the model and enable the trained probabilistic prediction model to improve prediction accuracy based on scene features.

[0222] The following section provides a more detailed description of the application scenarios corresponding to the resource recommendation method provided in this application embodiment. The resource recommendation method provided in this application embodiment can be applied to any application scenario involving resource recommendation, such as online social scenarios, online shopping scenarios, virtual game scenarios, AR scenarios, and VR scenarios; this application embodiment does not limit it in this regard.

[0223] I. Applied to virtual game scenarios.

[0224] Taking virtual game scenarios as an example, a game platform can connect to at least two game applications and recommend resources to players in at least two game applications through a recommendation system.

[0225] Players can log in to different game applications through a unified game account on the game platform. However, for different game applications or different time periods when players log in to game applications, the preferences of the same player may be different. Therefore, it is necessary to fully consider the characteristics of the scenario to determine the interaction probability between the player and the resource, so as to recommend corresponding resources based on the interaction probability and improve recommendation efficiency.

[0226] This application example illustrates the use of recommending consumer vouchers provided by brands integrated into game applications to players, with the interaction probability including the probability of players clicking on the consumer vouchers.

[0227] In the resource recommendation process, prediction data is acquired to predict the click probability of a player account and candidate resources. This prediction data includes attribute feature data and scene feature data. Attribute feature data indicates the attribute characteristics of the player account and candidate resources, while scene feature data indicates the scene characteristics of the interaction between the player account and candidate resources. Routing weights are determined based on the scene feature data, indicating the degree of influence of attribute feature data on the click probability. A probabilistic prediction model, a pre-trained machine learning model, is used to predict the click probability of a specified event based on the attribute feature data and the routing weights. If the click probability meets the recommendation requirements, candidate resources are recommended to the player account based on the click probability.

[0228] In some embodiments, the probabilistic prediction model includes at least two feature cross networks, each corresponding to a routing weight, which indicates the output weight of the feature cross network. Information features are extracted from the attribute feature data using the at least two feature cross networks in the probabilistic prediction model to obtain at least two feature cross results. Based on the routing weight corresponding to the i-th feature cross result and the i-th feature cross network, the output result of the i-th network is determined, where i is a positive integer. The click probability is predicted by the prediction network in the probabilistic prediction model based on the network output results corresponding to the at least two feature cross results.

[0229] Taking at least two feature cross-networks, including a low-order feature cross-network and a high-order feature cross-network, as an example, the low-order feature cross-network extracts low-order information features from the attribute feature data to obtain the low-order feature cross-network result; the high-order feature cross-network extracts high-order information features from the attribute feature data to obtain the high-order feature cross-network result; based on the routing weights corresponding to the i-th feature cross-network and the i-th feature cross-network among the at least two feature cross-network results, the i-th network output result is determined, including: determining the low-order network output result based on the low-order feature cross-network result and the low-order routing weights corresponding to the low-order feature cross-network, where the low-order routing weights are determined based on the network parameters of the low-order feature cross-network and the scene feature data; and determining the high-order network output result based on the high-order feature cross-network result and the high-order routing weights corresponding to the high-order feature cross-network, where the high-order routing weights are determined based on the network parameters of the high-order feature cross-network and the scene feature data.

[0230] In some embodiments, different dynamic routing methods need to be adopted for different feature cross-networks. Low-order routing weights are generated based on scene feature data through the dynamic weight layer corresponding to the low-order feature cross-network; high-order routing input data is obtained by concatenating scene feature data and attribute feature data; and high-order routing weights are generated based on the high-order routing input data through the dynamic weight layer corresponding to the high-order feature cross-network.

[0231] In some embodiments, the data needs to be vectorized first.

[0232] Indicatively, feature vectorization of attribute feature data yields attribute feature vectors, which include at least one of dense feature vectors, sparse feature vectors, and temporal feature vectors. Dense feature vectors are used to indicate continuous numerical features in the attribute feature data, sparse feature vectors are used to indicate categorical features in the attribute feature data, and temporal feature vectors are used to indicate temporal features in the attribute feature data.

[0233] Optionally, the low-order feature cross network includes low-order cross subnetworks and high-order cross subnetworks.

[0234] Information features are extracted based on dense and sparse feature vectors through a low-order crossover sub-network to obtain the low-order crossover result. The low-order crossover sub-network includes a first-order feature crossover network and a second-order feature crossover network. The first-order feature crossover network is used to quantify the independent influence of dense and sparse feature vectors on the click probability, and the second-order feature crossover network is used to quantify the joint influence of at least two dense feature vectors on the click probability. At least two of the attribute feature vectors are concatenated to obtain the high-order crossover input features. The high-order crossover input features are then processed layer by layer through the high-order crossover sub-network to obtain the high-order crossover result. Finally, the low-order crossover result and the high-order crossover result are fused to obtain the low-order feature crossover result.

[0235] Optionally, higher-order feature cross-networks include multi-expert cross-networks.

[0236] The dense feature vector, sparse feature vector, and temporal feature vector are concatenated to obtain the expert network input features. A general feature representation is extracted based on the expert network input features through a shared expert subnetwork in a multi-expert cross network. A task feature representation is extracted based on the expert network input features through a task expert subnetwork in a multi-expert cross network. The general feature representation and the task feature representation are fused to obtain the high-order feature cross result.

[0237] In some embodiments, the network outputs corresponding to at least two feature cross-results are concatenated to obtain fused features; the click probability is then predicted by a prediction network based on the fused features.

[0238] In some embodiments, account attribute data, resource attribute data, and context data are obtained as attribute feature data. The account attribute data is used to indicate the account data of the player account, the resource attribute data is used to indicate the resource attributes of the candidate resources, and the context data is used to indicate the environmental state of the player account. Program feature data and time feature data are obtained as scene feature data. The program feature data is used to indicate the program type of the candidate resources recommended to the player account, and the time feature data is used to indicate the recommendation time of the candidate resources recommended to the player account.

[0239] In some embodiments, the model training process includes: acquiring sample prediction data, which is used to predict the sample click probability of a sample account on a sample resource; the sample prediction data includes sample attribute data and sample scene data, and the sample prediction data corresponds to a reference label; determining sample routing weights based on the sample scene data, which are used to indicate the degree of influence of sample attribute data on the sample click probability; predicting the sample click probability of a sample account on a sample resource using a candidate probability prediction model based on the sample attribute data and the sample routing weights; and training the candidate probability prediction model based on the difference between the sample click probability and the reference label until the training requirements are met, thereby obtaining the probability prediction model.

[0240] In some embodiments, a first candidate resource whose click probability meets the recommendation requirements is obtained from the candidate resources; the first candidate resource is recommended to the player account in descending order of click probability, wherein the click probability of the first candidate resource is positively correlated with the recommendation display priority of the first candidate resource in the client where the player account is located, and the recommendation display priority is used to indicate the display order and visual prominence of the first candidate resource in the client where the player account is located.

[0241] It is worth noting that the above-mentioned click probabilities are merely illustrative examples. The resource recommendation method provided in this application embodiment can also recommend candidate resources based on at least one type of interaction probability, such as viewing probability, usage probability, and redemption probability. This application embodiment does not limit this.

[0242] II. Applied to multimedia playback scenarios.

[0243] Taking a multimedia playback scenario as an example, a multimedia playback platform can connect to at least two multimedia playback applications and recommend resources to viewers in at least two multimedia playback applications through a recommendation system.

[0244] Viewers can log in to different multimedia playback applications using a unified viewer account on the playback platform. However, for different multimedia playback applications, or for different time periods when a viewer logs into a multimedia playback application, the viewer's preferences for recommended resources may differ, even for the same viewer. Therefore, it is necessary to fully consider scenario feature parameters to determine the interaction probability between the viewer and the resources, and then recommend corresponding resources based on the interaction probability to improve recommendation efficiency.

[0245] This application example illustrates the use of recommending multimedia content from a multimedia playback application to a viewer, where the interaction probability includes the probability of the viewer viewing the multimedia content.

[0246] In the resource recommendation process, prediction data is acquired to predict the viewing probability of audience accounts and candidate resources. This prediction data includes attribute feature data and scene feature data. Attribute feature data indicates the attribute characteristics of audience accounts and candidate resources, while scene feature data indicates the scene characteristics of the interaction between audience accounts and candidate resources. Routing weights are determined based on the scene feature data, indicating the degree of influence of attribute feature data on the viewing probability. A probabilistic prediction model, a pre-trained machine learning model, is used to predict the viewing probability of audience accounts and candidate resources based on the attribute feature data and routing weights. If the viewing probability meets the recommendation requirements, candidate resources are recommended to the audience account based on the viewing probability.

[0247] It is worth noting that the above application scenarios are merely illustrative examples, and the embodiments of this application do not limit them.

[0248] Figure 12 This is a structural block diagram of a resource recommendation apparatus provided in an exemplary embodiment of this application, such as... Figure 12 As shown, the device includes the following parts:

[0249] The acquisition module 1210 is used to acquire prediction data, which is used to predict the interaction probability between the first account and the candidate resource. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resource, and the scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resource.

[0250] Processing module 1220 is used to determine routing weights based on the scene feature data, wherein the routing weights are used to indicate the degree of influence of the attribute feature data on the interaction probability;

[0251] The processing module 1220 is further configured to predict the interaction probability between the first account and the candidate resource based on the attribute feature data and the routing weight using a probability prediction model. The probability prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring.

[0252] The processing module 1220 is further configured to recommend the candidate resources to the first account based on the interaction probability when the interaction probability meets the recommendation requirements.

[0253] In some embodiments, the probabilistic prediction model includes at least two feature cross networks, each corresponding to a routing weight, and the routing weights are used to indicate the output weights of the feature cross networks.

[0254] The processing module 1220 is also used for:

[0255] Information features are extracted from the attribute feature data by at least two feature cross networks in the probability prediction model, respectively, to obtain at least two feature cross results;

[0256] Based on the routing weights corresponding to the i-th feature cross result and the i-th feature cross network among the at least two feature cross results, the output result of the i-th network is determined, where i is a positive integer;

[0257] The interaction probability is predicted by the prediction network in the probability prediction model based on the network output results corresponding to the at least two feature cross results.

[0258] In some embodiments, the at least two feature cross networks include a low-order feature cross network and a high-order feature cross network;

[0259] The processing module 1220 is also used for:

[0260] The low-order feature crossover network extracts low-order information features from the attribute feature data to obtain low-order feature crossover results.

[0261] The higher-order feature cross-network extracts higher-order information features from the attribute feature data to obtain higher-order feature cross-results.

[0262] Based on the low-order feature cross results and the low-order routing weights corresponding to the low-order feature cross network, the low-order network output results are determined. The low-order routing weights are determined based on the network parameters of the low-order feature cross network and the scene feature data.

[0263] Based on the higher-order feature cross results and the higher-order routing weights corresponding to the higher-order feature cross network, the output results of the higher-order network are determined. The higher-order routing weights are determined based on the network parameters of the higher-order feature cross network and the scene feature data.

[0264] In some embodiments, the processing module 1220 is further configured to:

[0265] The low-order routing weights are generated based on the scene feature data through the dynamic weight layer corresponding to the low-order feature cross network.

[0266] By concatenating the scene feature data and the attribute feature data, higher-order routing input data is obtained;

[0267] The higher-order routing weights are generated by the dynamic weight layer corresponding to the higher-order feature cross network based on the higher-order routing input data.

[0268] In some embodiments, the processing module 1220 is further configured to feature vectorize the attribute feature data to obtain an attribute feature vector, wherein the attribute feature vector includes at least one of dense feature vector, sparse feature vector and temporal feature vector, wherein the dense feature vector is used to indicate continuous numerical features in the attribute feature data, the sparse feature vector is used to indicate categorical features in the attribute feature data, and the temporal feature vector is used to indicate temporal features in the attribute feature data.

[0269] In some embodiments, the low-order feature cross network includes a low-order cross subnetwork and a high-order cross subnetwork;

[0270] The processing module 1220 is also used for:

[0271] The low-order crossover subnetwork extracts information features based on the dense feature vector and the sparse feature vector to obtain the low-order crossover result. The low-order crossover subnetwork includes a first-order feature crossover network and a second-order feature crossover network. The first-order feature crossover network is used to quantify the degree of independent influence of the dense feature vector and the sparse feature vector on the interaction probability. The second-order feature crossover network is used to quantify the degree of joint influence of at least two dense feature vectors on the interaction probability.

[0272] By concatenating at least two of the attribute feature vectors, a higher-order cross-input feature is obtained;

[0273] The higher-order cross input features are processed layer by layer through the higher-order cross sub-network to obtain higher-order cross sub-results;

[0274] The low-order feature cross result is obtained by fusing the low-order cross result and the high-order cross result.

[0275] In some embodiments, the high-order feature cross-network includes a multi-expert cross-network;

[0276] The processing module 1220 is also used for:

[0277] By concatenating the dense feature vector, the sparse feature vector, and the temporal feature vector, the input features of the expert network are obtained.

[0278] A general feature representation is extracted based on the input features of the expert network through a shared expert subnetwork in the multi-expert cross network;

[0279] Task feature representations are extracted based on the input features of the expert network through the task expert sub-network in the multi-expert cross network.

[0280] The higher-order feature cross result is obtained by fusing the general feature representation and the task feature representation.

[0281] In some embodiments, the processing module 1220 is further configured to:

[0282] By concatenating the network outputs corresponding to the at least two feature cross results, a fused feature is obtained;

[0283] The interaction probability is predicted by the prediction network based on the fusion features.

[0284] In some embodiments, the acquisition module 1210 is further configured to:

[0285] Account attribute data, resource attribute data, and context data are obtained as the attribute feature data. The account attribute data is used to indicate the account data of the first account, the resource attribute data is used to indicate the resource attributes of the candidate resources, and the context data is used to indicate the environmental state of the first account.

[0286] The program feature data and time feature data are obtained as the scene feature data. The program feature data is used to indicate the type of program that recommends the candidate resource to the first account, and the time feature data is used to indicate the time when the candidate resource is recommended to the first account.

[0287] In some embodiments, the acquisition module 1210 is further configured to acquire sample prediction data, which is used to predict the sample interaction probability of a sample account to a sample resource. The sample prediction data includes sample attribute data and sample scenario data, and the sample prediction data corresponds to a reference label.

[0288] The processing module 1220 is further configured to determine sample routing weights based on the sample scene data, wherein the sample routing weights are used to indicate the degree of influence of the sample attribute data on the sample interaction probability;

[0289] The processing module 1220 is further configured to predict the sample interaction probability of the sample account for the sample resource based on the sample attribute data and the sample routing weight using a candidate probability prediction model;

[0290] The processing module 1220 is further configured to train the candidate probability prediction model based on the difference between the sample interaction probability and the reference label until the training requirements are met, thereby obtaining the probability prediction model.

[0291] In some embodiments, the processing module 1220 is further configured to:

[0292] Obtain a first candidate resource whose interaction probability meets the recommendation requirements from the candidate resources;

[0293] The first candidate resource is recommended to the first account in descending order of the interaction probability, wherein the interaction probability of the first candidate resource is positively correlated with the recommendation display priority of the first candidate resource in the client where the first account is located, and the recommendation display priority is used to indicate the display order and visual prominence of the first candidate resource in the client where the first account is located.

[0294] In summary, the device provided in this application, by using the scene features of the interaction between the first account and the candidate resources to participate in the prediction of the interaction probability, can determine the routing weight based on the scene feature data, thereby adjusting the influence of attribute feature data on the interaction probability in different scenarios according to the scene features, improving the fit between the interaction probability predicted by the probability prediction model based on attribute feature data and the interaction scenario, improving the prediction accuracy of the interaction probability, and thus the recommendation efficiency of recommending candidate resources to the first account based on the interaction probability is high, which can adapt to differentiated recommendation scenarios.

[0295] It should be noted that the resource recommendation device provided in the above embodiments is only an example of 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 device can be divided into different functional modules to complete all or part of the functions described above.

[0296] Figure 13This illustration shows a structural block diagram of a terminal 1300 provided in an exemplary embodiment of this application. The terminal 1300 may be a smartphone, tablet computer, MP3 player, MP4 player, laptop computer, or desktop computer. The terminal 1300 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other names.

[0297] Typically, terminal 1300 includes a processor 1301 and a memory 1302.

[0298] Processor 1301 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1301 may be implemented using at least one hardware form selected from Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). Processor 1301 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1301 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 1301 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0299] Memory 1302 may include one or more computer-readable storage media, which may be non-transitory. Memory 1302 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in memory 1302 is used to store at least one instruction, which is executed by processor 1301 to implement the resource recommendation method provided in the method embodiments of this application.

[0300] In some embodiments, the terminal 1300 also includes other components, as those skilled in the art will understand. Figure 13 The structure shown does not constitute a limitation on terminal 1300 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0301] Embodiments of this application also provide a computer device that can be implemented as follows: Figure 1 The terminal or server shown. The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the resource recommendation method provided in the above-described method embodiments.

[0302] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the resource recommendation method provided in the above-described method embodiments.

[0303] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the resource recommendation method provided in the above-described method embodiments.

[0304] Optionally, the computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. The random access memory may include resistive random access memory (ReRAM) and dynamic random access memory (DRAM). The sequence numbers of the embodiments in this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0305] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0306] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A resource recommendation method, characterized in that, The method includes: Acquire prediction data, which is used to predict the probability of interaction between the first account and the candidate resource. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resource. The scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resource. The routing weight is determined based on the scene feature data, and the routing weight is used to indicate the degree of influence of the attribute feature data on the interaction probability. The probability of interaction between the first account and the candidate resource is predicted by a probabilistic prediction model based on the attribute feature data and the routing weight. The probabilistic prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring. If the interaction probability meets the recommendation requirements, the candidate resource is recommended to the first account based on the interaction probability.

2. The method according to claim 1, characterized in that, The probabilistic prediction model includes at least two feature cross networks, each of which corresponds to a routing weight, and the routing weight is used to indicate the output weight of the feature cross network. The step of predicting the interaction probability between the first account and the candidate resource based on the attribute feature data and the routing weight using a probabilistic prediction model includes: Information features are extracted from the attribute feature data by at least two feature cross networks in the probability prediction model, respectively, to obtain at least two feature cross results; Based on the routing weights corresponding to the i-th feature cross result and the i-th feature cross network among the at least two feature cross results, the output result of the i-th network is determined, where i is a positive integer; The interaction probability is predicted by the prediction network in the probability prediction model based on the network output results corresponding to the at least two feature cross results.

3. The method according to claim 2, characterized in that, The at least two feature cross networks include a low-order feature cross network and a high-order feature cross network; The step of extracting information features from the attribute feature data through at least two feature cross networks in the probability prediction model to obtain at least two feature cross results includes: The low-order feature crossover network extracts low-order information features from the attribute feature data to obtain low-order feature crossover results. The higher-order feature cross-network extracts higher-order information features from the attribute feature data to obtain higher-order feature cross-results. The step of determining the output result of the i-th network based on the routing weights corresponding to the i-th feature cross network among the at least two feature cross results includes: Based on the low-order feature cross results and the low-order routing weights corresponding to the low-order feature cross network, the low-order network output results are determined. The low-order routing weights are determined based on the network parameters of the low-order feature cross network and the scene feature data. Based on the higher-order feature cross results and the higher-order routing weights corresponding to the higher-order feature cross network, the output results of the higher-order network are determined. The higher-order routing weights are determined based on the network parameters of the higher-order feature cross network and the scene feature data.

4. The method according to claim 3, characterized in that, The step of determining the routing weight based on the scene feature data includes: The low-order routing weights are generated based on the scene feature data through the dynamic weight layer corresponding to the low-order feature cross network. By concatenating the scene feature data and the attribute feature data, higher-order routing input data is obtained; The higher-order routing weights are generated by the dynamic weight layer corresponding to the higher-order feature cross network based on the higher-order routing input data.

5. The method according to claim 3, characterized in that, The method further includes: The attribute feature data is vectorized to obtain attribute feature vectors, which include at least one of dense feature vectors, sparse feature vectors, and temporal feature vectors. The dense feature vectors are used to indicate continuous numerical features in the attribute feature data, the sparse feature vectors are used to indicate categorical features in the attribute feature data, and the temporal feature vectors are used to indicate temporal features in the attribute feature data.

6. The method according to claim 5, characterized in that, The low-order feature cross network includes a low-order cross sub-network and a high-order cross sub-network; The step of extracting low-order information features from the attribute feature data through the low-order feature cross network to obtain the low-order feature cross result includes: The low-order crossover subnetwork extracts information features based on the dense feature vector and the sparse feature vector to obtain the low-order crossover result. The low-order crossover subnetwork includes a first-order feature crossover network and a second-order feature crossover network. The first-order feature crossover network is used to quantify the degree of independent influence of the dense feature vector and the sparse feature vector on the interaction probability. The second-order feature crossover network is used to quantify the degree of joint influence of at least two dense feature vectors on the interaction probability. By concatenating at least two of the attribute feature vectors, a higher-order cross-input feature is obtained; The higher-order cross input features are processed layer by layer through the higher-order cross sub-network to obtain higher-order cross sub-results; The low-order feature cross result is obtained by fusing the low-order cross result and the high-order cross result.

7. The method according to claim 5, characterized in that, The higher-order feature cross-network includes a multi-expert cross-network; The step of extracting higher-order information features from the attribute feature data through the higher-order feature cross network to obtain the higher-order feature cross result includes: By concatenating the dense feature vector, the sparse feature vector, and the temporal feature vector, the input features of the expert network are obtained. A general feature representation is extracted based on the input features of the expert network through a shared expert subnetwork in the multi-expert cross network; Task feature representations are extracted based on the input features of the expert network through the task expert sub-network in the multi-expert cross network. The higher-order feature cross result is obtained by fusing the general feature representation and the task feature representation.

8. The method according to claim 2, characterized in that, The step of predicting the interaction probability through the prediction network in the probability prediction model based on the network output results corresponding to the at least two feature cross results includes: By concatenating the network outputs corresponding to the at least two feature cross results, a fused feature is obtained; The interaction probability is predicted by the prediction network based on the fusion features.

9. The method according to any one of claims 1 to 8, characterized in that, The acquisition of prediction data includes: Account attribute data, resource attribute data, and context data are obtained as the attribute feature data. The account attribute data is used to indicate the account data of the first account, the resource attribute data is used to indicate the resource attributes of the candidate resources, and the context data is used to indicate the environmental state of the first account. The program feature data and time feature data are obtained as the scene feature data. The program feature data is used to indicate the type of program that recommends the candidate resource to the first account, and the time feature data is used to indicate the time when the candidate resource is recommended to the first account.

10. The method according to any one of claims 1 to 8, characterized in that, The method further includes: Obtain sample prediction data, which is used to predict the probability of sample interaction between sample accounts and sample resources. The sample prediction data includes sample attribute data and sample scenario data, and the sample prediction data corresponds to reference tags. The sample routing weight is determined based on the sample scenario data, and the sample routing weight is used to indicate the degree of influence of the sample attribute data on the sample interaction probability. The candidate probability prediction model predicts the sample interaction probability of the sample account for the sample resource based on the sample attribute data and the sample routing weight; The candidate probability prediction model is trained based on the difference between the sample interaction probability and the reference label until the training requirements are met, thus obtaining the probability prediction model.

11. The method according to any one of claims 1 to 8, characterized in that, When the interaction probability meets the recommendation requirements, recommending the candidate resources to the first account based on the interaction probability includes: Obtain a first candidate resource whose interaction probability meets the recommendation requirements from the candidate resources; The first candidate resource is recommended to the first account in descending order of the interaction probability, wherein the interaction probability of the first candidate resource is positively correlated with the recommendation display priority of the first candidate resource in the client where the first account is located, and the recommendation display priority is used to indicate the display order and visual prominence of the first candidate resource in the client where the first account is located.

12. A resource recommendation device, characterized in that, The device includes: An acquisition module is used to acquire prediction data, which is used to predict the probability of interaction between a first account and a candidate resource. The prediction data includes attribute feature data and scene feature data. The attribute feature data is used to indicate the attribute features of the first account and the candidate resource, and the scene feature data is used to indicate the scene features of the interaction between the first account and the candidate resource. The processing module is used to determine the routing weight based on the scene feature data, wherein the routing weight is used to indicate the degree of influence of the attribute feature data on the interaction probability; The processing module is further configured to predict the interaction probability between the first account and the candidate resource based on the attribute feature data and the routing weight using a probability prediction model. The probability prediction model is a pre-trained machine learning model used to predict the probability of a specified event occurring. The processing module is further configured to recommend the candidate resources to the first account based on the interaction probability if the interaction probability meets the recommendation requirements.

13. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the resource recommendation method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The storage medium stores at least one computer program, which is loaded and executed by a processor to implement the resource recommendation method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the resource recommendation method as described in any one of claims 1 to 11.

Images