Resource recommendation method and device, electronic equipment and storage medium

Abstract

The invention discloses a resource recommendation method and device, electronic equipment and a storage medium, and relates to the technical field of computers, in particular to the field of artificial intelligence such as deep learning. According to the specific implementation scheme, the method comprises the steps of obtaining a multi-target prediction result of each candidate resource, comprehensive user features of a target user and key user features; performing feature extraction on the key user features by adopting a first neural network in a target evolutionary strategy model to obtain a first feature vector; processing the comprehensive user features and the first feature vector by adopting a second neural network in a target evolutionary strategy model to obtain personalized fusion parameters of each target corresponding to the target user; determining resource scores of the candidate resources according to the personalized fusion parameters and the multi-target prediction result; and according to the resource score of each candidate resource, carrying out resource recommendation to the target user.

Description

Technical Field

[0001] The present application relates to the field of computer technology, in particular to artificial intelligence fields such as deep learning, and specifically to a resource recommendation method, device, electronic device and storage medium. Background Art

[0002] With the development of artificial intelligence, in order to improve user experience, resource recommendation systems can recommend resources to users based on user behavior data. Summary of the Invention

[0003] This application provides a resource recommendation method, device, electronic device, and storage medium. The details are as follows:

[0004] According to one aspect of the present application, a resource recommendation method is provided, comprising:

[0005] Obtain multi-objective prediction results for each candidate resource, comprehensive user characteristics of target users, and key user characteristics;

[0006] Using a first neural network in a target evolution strategy model, extracting features from the key user features to obtain a first feature vector;

[0007] Using the second neural network in the target evolution strategy model, the comprehensive user characteristics and the first feature vector are processed to obtain personalized fusion parameters of each target corresponding to the target user;

[0008] Determining a resource score of the candidate resource according to the personalized fusion parameter and the multi-objective prediction result;

[0009] Recommend resources to the target user based on the resource score of each candidate resource.

[0010] According to another aspect of the present application, a resource recommendation device is provided, comprising:

[0011] An acquisition module is used to obtain the multi-objective prediction results of each candidate resource, the comprehensive user characteristics of the target user, and the key user characteristics;

[0012] A first feature processing module is used to extract the key user features using the first neural network in the target evolution strategy model to obtain a first feature vector;

[0013] a second feature processing module, configured to process the comprehensive user features and the first feature vector using a second neural network in the target evolution strategy model to obtain personalized fusion parameters for each target corresponding to the target user;

[0014] a determination module, configured to determine a resource score of the candidate resource based on the personalized fusion parameter and the multi-objective prediction result;

[0015] The recommendation module is used to recommend resources to the target user based on the resource scores of the candidate resources.

[0016] According to another aspect of the present application, an electronic device is provided, including:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the above embodiment.

[0020] According to another aspect of the present application, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method according to the above embodiment.

[0021] According to another aspect of the present application, a computer program product is provided, including a computer program, which implements the steps of the method described in the above embodiment when executed by a processor.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present application.

[0024] Figure 1 A flowchart of a resource recommendation method provided in one embodiment of the present application;

[0025] Figure 2 A flowchart of a resource recommendation method provided in another embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of a target evolution strategy model provided in an embodiment of the present application;

[0027] Figure 4 A flowchart of a resource recommendation method provided in another embodiment of the present application;

[0028] Figure 5A schematic diagram of the training process of a target evolution strategy model provided in an embodiment of the present application;

[0029] Figure 6 A schematic diagram of the structure of a resource recommendation device provided in one embodiment of the present application;

[0030] Figure 7 4 is a block diagram of an electronic device used to implement the resource recommendation method of an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following description of exemplary embodiments of the present application is made in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations and do not violate public order and good morals.

[0033] The resource recommendation method, device, electronic device, and storage medium of the embodiments of the present application are described below with reference to the accompanying drawings.

[0034] Figure 1 A flowchart of a resource recommendation method provided in one embodiment of the present application.

[0035] The resource recommendation method of the embodiment of the present application can be executed by the resource recommendation device of the embodiment of the present application, and the device can be configured in an electronic device.

[0036] Among them, the electronic device can be any device with computing capabilities, such as a personal computer, mobile terminal, server, etc. The mobile terminal can be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, and other hardware devices with various operating systems, touch screens and / or display screens.

[0037] like Figure 1 As shown, the resource recommendation method includes:

[0038] Step 101: Obtain multi-objective prediction results of each candidate resource, comprehensive user characteristics and key user characteristics of the target user.

[0039] The resources to be recommended in this application may be, for example, videos, pictures, or commodities, etc., without limitation.

[0040] In this application, a multi-objective model can be used to perform multi-objective prediction on each candidate resource to obtain a multi-objective prediction result. The multi-objective can be related to the resource category. Taking video resources as an example, the multi-objective can include but is not limited to satisfaction, completion rate, playback time, completion, fast scrolling, etc.

[0041] In this application, the comprehensive user characteristics and key user characteristics of the target user can be obtained based on the target user's historical behavior data on resources, current scenario information, etc.

[0042] The number of user features included in the comprehensive user features is greater than a preset threshold, and the key user features may be key user features in the comprehensive user features, which can better reflect the personalized features of the user.

[0043] As an example, taking video resources as an example, the key user characteristics are shown in Table 1, and the comprehensive user characteristics are shown in Table 2.

[0044] Table 1

[0045] Feature Classification Feature Examples Scene characteristics Is it a rest day? Is there a rest time? Request Features Brush Demographic characteristics User activity

[0046] Table 2

[0047]

[0048]

[0049] It can be seen that according to the feature examples corresponding to each feature category in Table 2, Table 2 contains 11 user features. According to the feature examples corresponding to each feature category in Table 1, there are 4 user features in Table 1. The number of user features in Table 2 is greater than the number of user features in Table 1.

[0050] Step 102: Using the first neural network in the target evolution strategy model, extract key user features to obtain a first feature vector.

[0051] In this application, the target evolution strategy model may include a first neural network and a second neural network. The first neural network acts as an independent network to learn key user characteristics such as crowd characteristics and scene characteristics, and is used to extract key user characteristics. The second neural network is the main network of the target evolution strategy model.

[0052] Exemplarily, the name of the first neural network can be called POSO (Personalized Cold Start) network.

[0053] Exemplarily, the first neural network may include a hidden layer, and the hidden layer in the first neural network may be used to extract key user features to obtain a first feature vector.

[0054] For example, the dimension of the key user feature is 9, and a 16-dimensional feature vector can be obtained through the hidden layer.

[0055] Step 103 : Using the second neural network in the target evolution strategy model, the comprehensive user features and the first feature vector are processed to obtain personalized fusion parameters of each target corresponding to the target user.

[0056] In the present application, a second neural network may be used to extract comprehensive user features, and based on the extracted feature vector and the first feature vector, personalized fusion parameters of each target corresponding to the target user may be obtained.

[0057] Step 104 : Determine the resource score of the candidate resource based on the personalized fusion parameters and the multi-objective prediction results.

[0058] In this application, the multi-target prediction results of any candidate resource can be weighted and fused according to the personalized fusion parameters of each target to obtain the resource score of any candidate resource.

[0059] Step 105: Recommend resources to the target user based on the resource score of each candidate resource.

[0060] In this application, the candidate resources may be sorted in descending order of resource scores, and a preset number of resources in the sorting results may be recommended to the target user.

[0061] In the implementation of this application, the feature vectors of key user features extracted by the first neural network in the target evolution strategy model are input into the second neural network in the target evolution strategy model and processed in combination with the comprehensive user features, so that the model can make personalized responses to each user at the model structure level, thereby improving the accuracy of personalized fusion parameters, and thus making resource recommendations based on personalized fusion parameters, thereby improving the accuracy of resource recommendations.

[0062] Figure 2 A flowchart of a resource recommendation method provided in another embodiment of the present application.

[0063] like Figure 2 As shown, the resource recommendation method includes:

[0064] Step 201 : Obtain multi-objective prediction results of each candidate resource, comprehensive user characteristics and key user characteristics of the target user.

[0065] Step 202 : Using the first neural network in the target evolution strategy model, extract key user features to obtain a first feature vector.

[0066] In the present application, steps 201 and 202 can be implemented in any of the embodiments of the present application, and therefore will not be described in detail here.

[0067] Step 203: Use the first sub-network in the second neural network to extract the comprehensive user features to obtain a second feature vector.

[0068] In the present application, the second neural network may be a multi-layer network, and the first sub-network in the second neural network may be used to perform feature extraction on the comprehensive user features to obtain a second feature vector.

[0069] For example, the first sub-network can be a hidden layer, the dimension of the comprehensive user features is 49 dimensions, the weight matrix size of the hidden layer is 49*16, then the dimension of the second feature vector is 16 dimensions.

[0070] Step 204: fuse the first eigenvector and the second eigenvector to obtain a third eigenvector.

[0071] In the present application, the first eigenvector may be used as a weight to weight the second eigenvector to obtain a third eigenvector.

[0072] For example, the first eigenvector is 16-dimensional, the second eigenvector is also 16-dimensional, and the third eigenvector obtained after fusion is also 16-dimensional.

[0073] Step 205: Based on the third eigenvector, a second sub-network in the second neural network is used for decoding to obtain personalized fusion parameters of each target.

[0074] As a possible implementation method, the second sub-network in the second neural network can be used for decoding based on the third eigenvector, and the second neural network outputs personalized fusion parameters for each target.

[0075] As another possible implementation method, based on the third eigenvector, a second sub-network is used for decoding, and the second neural network outputs the fusion parameter adjustment amount of each target corresponding to the target user. According to the fusion parameter adjustment amount of any target and the basic fusion parameter of the target, the personalized fusion parameter of the target is determined.

[0076] Exemplarily, the fusion parameter adjustment amount of any target may be used to adjust the basic fusion parameters of the target to obtain the personalized fusion parameters of the target.

[0077] As an example, the personalized fusion parameter can be calculated by the following formula (1):

[0078] p′=p * (1+Δp) (1)

[0079] Among them, p′ represents the personalized fusion parameter; p * represents the basic fusion parameter; Δp represents the adjustment amount of the fusion parameter.

[0080] Therefore, the amount of calculation of the model can be reduced by decoding and outputting the fusion parameter adjustment amount of each target through the second neural network.

[0081] Exemplarily, the third eigenvector of the second sub-network may be used for direct decoding to obtain the fusion parameter adjustment amount of each target corresponding to the target user.

[0082] Exemplarily, the second neural network may further include a third sub-network, which may first be used to extract features from the second eigenvector to obtain a third eigenvector, and then the second sub-network may be used to decode the fourth eigenvector to obtain the fusion parameter adjustment amounts of each target corresponding to the target user.

[0083] For example, the third sub-network may be a hidden layer, and the hidden layer is used to extract features from the second feature vector.

[0084] As a result, the second neural network includes a multi-layer sub-network for feature extraction, and the parameter space is significantly expanded, thereby improving the model's expressiveness and further improving the accuracy of personalized fusion parameters.

[0085] As an example, Figure 3 In the structure of the target evolution strategy model shown, the target evolution strategy model includes a personalized cold start network and a main network. The personalized cold start network includes a first hidden layer, and the main network includes a second hidden layer B and a third hidden layer. The first hidden layer transforms the 16-dimensional key user features through the weight matrix Wp to obtain a 16-dimensional output. The second hidden layer transforms the 49-dimensional comprehensive user features through the weight matrix W0 to obtain a 16-dimensional output. The output of the first hidden layer is fused with the output of the second hidden layer, and the fusion result is processed by the third hidden layer. The output of the third hidden layer is then transformed by the weight matrix W1 to obtain a 10-dimensional output result.

[0086] It should be noted that the target evolution strategy model outputs personalized fusion parameters and also integrates parameter adjustment amounts, which may be related to model training, model internal processing, etc., and is not limited to this.

[0087] In addition, in the present application, the target evolution strategy model uses a multi-layer sub-network in the second neural network for feature extraction to process the first feature vector and comprehensive user features, and output the method of fusion parameter adjustment. This is similar to the method in which the target evolution strategy model uses a multi-layer sub-network in the second neural network for feature extraction to process the first feature vector and comprehensive user features, and output personalized fusion parameters, so it will not be repeated here.

[0088] Step 206 : Determine the resource score of the candidate resource based on the personalized fusion parameters and the multi-objective prediction results.

[0089] Step 207: Recommend resources to the target user based on the resource score of each candidate resource.

[0090] In the present application, steps 206 and 207 can be implemented in any of the embodiments of the present application, so they will not be described in detail here.

[0091] In an embodiment of the present application, the first sub-network in the second neural network is used to extract the comprehensive user features to obtain a second feature vector, and the second feature vector is fused with the first feature vector to obtain a third feature vector. Based on the third feature vector, the second sub-network in the second neural network is used for decoding to obtain personalized fusion parameters. Thus, by fusing the feature vector extracted from the comprehensive user features with the feature vector extracted from the key user features, and decoding based on the fused feature vector, the accuracy of the personalized fusion parameters can be improved.

[0092] Figure 4 A flowchart of a resource recommendation method provided in another embodiment of the present application.

[0093] like Figure 4 As shown, the resource recommendation method includes:

[0094] Step 401 : Obtain multi-objective prediction results of each candidate resource, comprehensive user characteristics and key user characteristics of the target user.

[0095] Step 402 : Using the first neural network in the target evolution strategy model, extract key user features to obtain a first feature vector.

[0096] Step 403 : Using the second neural network in the target evolution strategy model, the comprehensive user features and the first feature vector are processed to obtain personalized fusion parameters of each target corresponding to the target user.

[0097] In the present application, steps 401 to 403 can be implemented in any of the embodiments of the present application, and therefore will not be described in detail here.

[0098] Step 404: Adjust the personalized fusion parameters according to the attribute information of the candidate resources to obtain adjusted personalized fusion parameters.

[0099] The attribute information of the candidate resource may include but is not limited to the resource type, content category, resource tag, etc. of the candidate resource.

[0100] For example, taking video resources as an example, content categories may include but are not limited to technology, news, entertainment, etc., and resource tags may include but are not limited to "academic paper interpretation", "series explanation", etc.

[0101] In the present application, the adjustment amount of the personalized fusion parameter of each target can be determined according to the attribute information of the candidate resource, and the personalized fusion parameter can be adjusted according to the adjustment amount.

[0102] Since resources of different content categories may have different requirements for the same target, based on this, as a possible implementation method, the attribute information of the candidate resource includes the content category of the candidate resource, and the importance weight of each target corresponding to the content category of the candidate resource can be determined. According to the importance weight of each target, the personalized fusion parameters of each target are adjusted to obtain the adjusted personalized fusion parameters of each target.

[0103] For example, the personalized fusion parameters corresponding to a set number of targets with the highest importance weights can be adjusted, and the personalized fusion parameters of other targets can be reduced according to their importance weights. For example, the smaller the importance weight of a target, the greater the adjustment amount of the target's importance weight.

[0104] For example, for science and technology videos, since users may tend to watch professional content in depth, the importance weights of [playback time] and [completion rate] are relatively high. Therefore, the personalized fusion parameters of [playback time] and [completion rate] can be increased, and the personalized fusion parameters of other targets can be reduced or remain unchanged. For news videos, since it is necessary to balance information timeliness and content quality, the importance weights of [fast scrolling] and [satisfaction] are relatively high. Therefore, the personalized fusion parameters of [fast scrolling rate] and [satisfaction] can be increased, and the personalized fusion parameters of other targets can be reduced or remain unchanged.

[0105] Therefore, the personalized fusion parameters of each target can be adjusted based on the importance weight of each target corresponding to the content category of each candidate resource, so that the personalized fusion parameters can be adaptively adjusted based on the content category of the candidate resources, which can further improve the accuracy of the personalized fusion parameters and thus improve the accuracy of resource recommendations.

[0106] Since resources with different resource labels may have different requirements for the same target, based on this, as a possible implementation method, the key target corresponding to the resource label is determined from each target, and the personalized fusion parameters corresponding to the key target are adjusted to obtain the adjusted personalized fusion parameters corresponding to the key target.

[0107] Exemplarily, the key target corresponding to the resource tag of the candidate resource may be determined according to a preset mapping relationship between the resource tag and the key target.

[0108] Exemplarily, the personalized fusion parameters may be adjusted according to preset parameter adjustment amounts corresponding to key objectives to obtain adjusted personalized fusion parameters.

[0109] Taking video resources as an example, the video tag of a certain video includes "academic paper interpretation". Since users are likely to watch it in depth, the personalized fusion parameter of [playback time] can be increased, and the personalized fusion parameters of other targets can be reduced or remain unchanged.

[0110] Therefore, the personalized fusion parameters of the key targets can be adjusted based on the key targets corresponding to the resource tags of the candidate resources, thereby increasing the weight of the key targets in the multi-target fusion and improving the accuracy of resource recommendation.

[0111] Step 405 : Based on the adjusted personalized fusion parameters, weighted fusion is performed on the multi-objective prediction results of the candidate resources to obtain a resource score of the candidate resources.

[0112] In this application, the multi-objective prediction results of any candidate resource can be weightedly fused according to the adjusted personalized fusion parameters corresponding to the candidate resource to obtain the resource score of the candidate resource.

[0113] Step 406: Recommend resources to the target user based on the resource score of each candidate resource.

[0114] In this application, step 406 can be implemented in any of the embodiments of this application, so it will not be described in detail here.

[0115] In an embodiment of the present application, since the attribute information of different candidate resources may be different, adjusting the personalized fusion parameters according to the attribute information of the candidate resources can improve the accuracy of the personalized fusion parameters of the target user, thereby making resource recommendations based on the adjusted personalized fusion parameters, which can improve the accuracy of resource recommendations.

[0116] In one embodiment of the present application, the target evolution strategy model can be trained in the following manner:

[0117] The current model parameters of the target evolution strategy model can be perturbed to obtain multiple sets of candidate model parameters. The reward value corresponding to each set of candidate model parameters can be determined based on the user posterior data corresponding to each set of candidate model parameters. Based on the reward value of each set of candidate model parameters, the seed model parameters can be determined from the multiple sets of candidate model parameters. The seed model parameters can then be further perturbed to obtain multiple sets of candidate model parameters. The seed model parameters can then be selected from the candidate model parameters until the training end conditions are met to obtain the final model parameters of the target evolution strategy model.

[0118] For example, the candidate model parameters with the highest reward value can be used as seed model parameters.

[0119] Exemplarily, multiple groups of disturbance parameters may be determined based on current model parameters, and the current model parameters may be disturbed according to the multiple groups of disturbance parameters to obtain multiple groups of candidate model parameters.

[0120] For example, based on the current model parameters and covariance matrix, multiple sets of disturbance parameters can be obtained, thereby determining the globally optimal personalized fusion parameters corresponding to each target through the covariance matrix evolution method.

[0121] Exemplarily, for any set of candidate model parameters, the set of candidate model parameters can be used to determine the personalized parameters corresponding to the exploration group users based on the comprehensive user characteristics and key user characteristics corresponding to the exploration group users corresponding to the set of candidate model parameters, and resources are recommended to the exploration group users based on the personalized fusion parameters corresponding to the exploration group users. The posterior data of the exploration group users on the recommended resources is obtained, and the reward value corresponding to the candidate model parameters is determined based on the posterior data.

[0122] Among them, each group of candidate model parameters has a corresponding exploration group. In different training rounds, multiple exploration groups can be re-determined. For example, the entire user traffic can be divided into multiple exploration groups by performing hash bucketing.

[0123] Among them, the process of determining the personalized parameters corresponding to the exploration group users by using any set of candidate model parameters based on the comprehensive user characteristics and key user characteristics corresponding to the exploration group users corresponding to any set of candidate model parameters is similar to the method of determining the personalized fusion parameters corresponding to the target user by using the target evolution strategy model based on the comprehensive user characteristics and key user characteristics corresponding to the target user in the above embodiment, so it will not be repeated here.

[0124] Among them, the method of recommending resources to exploration group users based on the personalized fusion parameters corresponding to the exploration group users is similar to the method of recommending resources to target users based on the personalized fusion parameters corresponding to the target users in the above embodiment, so it will not be repeated here.

[0125] For example, based on the basic fusion parameters, resource recommendations can be made to the entire user traffic. This means that the entire user traffic is used as a control group. Based on the posterior data corresponding to the control group and the user posterior data corresponding to any set of candidate model parameters, the loss or gain of the resource evaluation indicators corresponding to the candidate model parameters relative to the resource evaluation indicators of the control group for the same user is determined, thereby determining the reward value of the set of candidate model parameters. The resource evaluation indicators are determined based on the posterior data, such as video resources, and the number of videos viewed by users, the number of videos in which users have interacted, etc.

[0126] In order to facilitate the understanding of the training method of the target evolution strategy model of this application, the following Figure 5 To explain, Figure 5 A schematic diagram of the training process of a target evolution strategy model provided in an embodiment of the present application.

[0127] like Figure 5 As shown, taking the daily update of model parameters as an example, on day T-1, part of the user traffic or all of the user traffic can be divided into 10 exploration groups, and the model parameters can be perturbed differently by using the covariance matrix-based evolution method to obtain a set of candidate model parameters corresponding to the 10 exploration groups on day T-1, that is, the candidate model parameters W_(1,T-1) corresponding to exploration group 1, the candidate model parameters W_(2,T-1) corresponding to exploration group 2, ..., and the candidate model parameters W_(10,T-1) corresponding to exploration group 10 are obtained.

[0128] For each exploration group, we can use its corresponding candidate model parameters to recommend resources to users in the exploration group online, obtaining the corresponding posterior data for each exploration group, namely, the posterior data for exploration group 1 (1, T-1), the posterior data for exploration group 2 (2, T-1), ..., and the posterior data for exploration group 10 (10, T-1). For users in the control group, we can use the basic fusion parameters to recommend resources.

[0129] Based on the posterior data of each exploration group and the posterior data of the control page, the reward values ​​of the candidate model parameters corresponding to the 10 exploration groups on day T-1 are calculated, and the candidate model parameters corresponding to the exploration group with the highest reward value are selected from the candidate model parameters corresponding to the 10 exploration groups on day T-1 as the seed model parameters W_best(T-1). The candidate model parameters corresponding to the 10 exploration groups on day T are derived from the seed model parameters W_best(T-1) based on the covariance matrix evolution method. 10 Is the disturbance parameter. According to the disturbance parameter σ1, the seed model parameters are disturbed to obtain a set of candidate model parameters W_(1,T) on day T, and according to the disturbance parameter σ2, the seed model parameters are disturbed to obtain a set of candidate model parameters W_(2,T) on day T, ..., according to the disturbance parameter σ 10 The seed model parameters are perturbed to obtain a set of candidate model parameters W_(10,T) for day T.

[0130] Afterwards, the candidate model parameters W_(1,T), W_(2,T), …, W_(10,T) corresponding to the 10 exploration groups on day T can be configured online. Based on the model-estimated fusion parameter adjustment amount, personalized fusion parameters are obtained based on the fusion parameter adjustment amount and the baseline fusion parameters. Based on the personalized fusion parameters, resources are recommended online to the users of the corresponding exploration groups, so as to obtain the posterior data of the users of the 10 exploration groups on day T, so as to select the seed model parameters for day T and prepare for the training on the next day until the training end conditions are met, such as the model parameters are stable, and the final model parameters of the target evolution strategy model are obtained.

[0131] Therefore, training the model parameters of the first neural network and the model parameters of the second neural network in the target evolution strategy model together can improve the training efficiency. The first neural network in the target evolution strategy model learns personalized signals as an independent network, and its output is input into the second neural network for processing, thereby improving the model's expression ability and the ability to capture personalized features, and realizing the model's personalized response to each user at the model structure level.

[0132] In order to implement the above embodiment, the embodiment of the present application further proposes a resource recommendation device. Figure 6 A schematic diagram of the structure of a resource recommendation device provided in one embodiment of the present application.

[0133] like Figure 6 As shown, the resource recommendation device 600 includes:

[0134] An acquisition module 610 is used to obtain the multi-objective prediction results of each candidate resource, the comprehensive user characteristics and key user characteristics of the target user;

[0135] A first feature processing module 620 is configured to extract the key user features using a first neural network in a target evolution strategy model to obtain a first feature vector;

[0136] A second feature processing module 630 is configured to process the comprehensive user features and the first feature vector using the second neural network in the target evolution strategy model to obtain personalized fusion parameters for each target corresponding to the target user;

[0137] A determination module 640 is configured to determine a resource score of the candidate resource based on the personalized fusion parameter and the multi-objective prediction result;

[0138] The recommendation module 650 is configured to recommend resources to the target user based on the resource scores of the candidate resources.

[0139] Optionally, the second feature processing module 630 is configured to:

[0140] Using the first sub-network in the second neural network, extracting features from the comprehensive user features to obtain a second feature vector;

[0141] Fusing the first eigenvector and the second eigenvector to obtain a third eigenvector;

[0142] Based on the third eigenvector, decoding is performed using the second subnetwork in the second neural network to obtain personalized fusion parameters for each target.

[0143] Optionally, the second feature processing module 630 is configured to:

[0144] Based on the third eigenvector, decoding is performed using the second sub-network to obtain a fusion parameter adjustment amount of each target corresponding to the target user;

[0145] The personalized fusion parameters of each target are determined according to the fusion parameter adjustment amount and the basic fusion parameters of each target.

[0146] Optionally, the second feature processing module 630 is configured to:

[0147] Using the third subnetwork in the second neural network, perform feature extraction on the third eigenvector to obtain a fourth eigenvector;

[0148] The second sub-network is used to decode the fourth eigenvector to obtain the fusion parameter adjustment amount.

[0149] Optionally, the target evolution strategy model is trained in the following manner:

[0150] Perturbing the current model parameters of the target evolution strategy model to obtain multiple sets of candidate model parameters;

[0151] Determining a reward value corresponding to the candidate model parameter based on user posterior data corresponding to the candidate model parameter;

[0152] determining seed model parameters from the multiple sets of candidate model parameters according to the reward value;

[0153] The seed model parameters are continuously perturbed to obtain the final model parameters of the target evolution strategy model.

[0154] Optionally, the determination module 640 is configured to:

[0155] Adjusting the personalized fusion parameters according to the attribute information of the candidate resources to obtain adjusted personalized fusion parameters;

[0156] Based on the adjusted personalized fusion parameters, weighted fusion is performed on the multi-objective prediction results of the candidate resources to obtain a resource score of the candidate resources.

[0157] Optionally, the attribute information includes a content category of the candidate resource. The determination module 640 is configured to:

[0158] Determining the importance weight of each of the objectives corresponding to the content category;

[0159] The personalized fusion parameter is adjusted according to the importance weight to obtain an adjusted personalized fusion parameter.

[0160] Optionally, the attribute information includes a resource tag of the candidate resource. The determination module 640 is configured to:

[0161] Determine the key target corresponding to the resource tag from the targets;

[0162] The personalized fusion parameters corresponding to the key target are adjusted to obtain adjusted personalized fusion parameters corresponding to the key target.

[0163] It should be noted that the explanations of the aforementioned resource recommendation method embodiment are also applicable to the resource recommendation device of this embodiment, and therefore will not be repeated here.

[0164] In the implementation of this application, the feature vectors of key user features extracted by the first neural network in the target evolution strategy model are input into the second neural network in the target evolution strategy model and processed in combination with the comprehensive user features, so that the model can make personalized responses to each user at the model structure level, thereby improving the accuracy of personalized fusion parameters, and thus making resource recommendations based on personalized fusion parameters, thereby improving the accuracy of resource recommendations.

[0165] According to an embodiment of the present application, the present application also provides an electronic device, a readable storage medium and a computer program product.

[0166] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0167] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 702 or a computer program loaded from a storage unit 708 into a RAM (Random Access Memory) 703. Various programs and data required for the operation of the device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An I / O (Input / Output) interface 705 is also connected to the bus 704.

[0168] Various components in device 700 are connected to I / O interface 705, including an input unit 706, such as a keyboard, mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, optical disk, etc.; and a communication unit 709, such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0169] The computing unit 701 can be various general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units that run machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the resource recommendation method described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to execute the resource recommendation method in any other appropriate manner (for example, by means of firmware).

[0170] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System on Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0171] The program code for implementing the methods of the present application can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow charts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0172] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or apparatus. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0173] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0174] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0175] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. The client-server relationship is established by computer programs running on the respective computers and establishing a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service ecosystem that addresses the management difficulties and poor scalability of traditional physical hosts and VPS services. The server may also be a server in a distributed system or a server integrated with blockchain.

[0176] According to an embodiment of the present application, the present application further provides a computer program product, which, when an instruction processor in the computer program product is executed, executes the resource recommendation method proposed in the above embodiment of the present application.

[0177] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this application can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this application can be achieved. This is not a limitation herein.

[0178] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. A resource recommendation method, comprising: Obtain multi-objective prediction results for each candidate resource, comprehensive user characteristics of target users, and key user characteristics; Using a first neural network in a target evolution strategy model, extracting features from the key user features to obtain a first feature vector; Using the second neural network in the target evolution strategy model, the comprehensive user characteristics and the first feature vector are processed to obtain personalized fusion parameters of each target corresponding to the target user; Determining a resource score of the candidate resource according to the personalized fusion parameter and the multi-objective prediction result; Recommend resources to the target user based on the resource score of each candidate resource.

2. The method according to claim 1, wherein The second neural network in the target evolution strategy model is used to process the comprehensive user characteristics and the first feature vector to obtain personalized fusion parameters of each target corresponding to the target user, including: Using the first sub-network in the second neural network, extracting features from the comprehensive user features to obtain a second feature vector; Fusing the first eigenvector and the second eigenvector to obtain a third eigenvector; Based on the third eigenvector, decoding is performed using the second subnetwork in the second neural network to obtain personalized fusion parameters for each target.

3. The method according to claim 2, wherein: The decoding using the second subnetwork in the second neural network based on the third eigenvector to obtain the personalized fusion parameters of each target includes: Based on the third eigenvector, decoding is performed using the second sub-network to obtain a fusion parameter adjustment amount of each target corresponding to the target user; The personalized fusion parameters of each target are determined according to the fusion parameter adjustment amount and the basic fusion parameters of each target.

4. The method according to claim 3, wherein: The decoding using the second sub-network based on the third eigenvector to obtain the fusion parameter adjustment amount of each target corresponding to the target user includes: Using the third subnetwork in the second neural network, perform feature extraction on the third eigenvector to obtain a fourth eigenvector; The second sub-network is used to decode the fourth eigenvector to obtain the fusion parameter adjustment amount.

5. The method according to claim 1, wherein The target evolution strategy model is trained in the following way: Perturbing the current model parameters of the target evolution strategy model to obtain multiple sets of candidate model parameters; Determining a reward value corresponding to the candidate model parameter based on user posterior data corresponding to the candidate model parameter; determining seed model parameters from the multiple sets of candidate model parameters according to the reward value; The seed model parameters are continuously perturbed to obtain the final model parameters of the target evolution strategy model.

6. The method according to any one of claims 1 to 5, wherein The determining of the resource score of the candidate resource according to the personalized fusion parameter and the multi-objective prediction result includes: Adjusting the personalized fusion parameters according to the attribute information of the candidate resources to obtain adjusted personalized fusion parameters; Based on the adjusted personalized fusion parameters, weighted fusion is performed on the multi-objective prediction results of the candidate resources to obtain a resource score of the candidate resources.

7. The method according to claim 6, wherein: The attribute information includes a content category of the candidate resource, and adjusting the personalized fusion parameter according to the attribute information of the candidate resource to obtain the adjusted personalized fusion parameter includes: Determining the importance weight of each of the objectives corresponding to the content category; The personalized fusion parameter is adjusted according to the importance weight to obtain an adjusted personalized fusion parameter.

8. The method of claim 6, wherein: The attribute information includes a resource tag of the candidate resource, and adjusting the personalized fusion parameter according to the attribute information of the candidate resource to obtain the adjusted personalized fusion parameter includes: Determine the key target corresponding to the resource tag from the targets; The personalized fusion parameters corresponding to the key target are adjusted to obtain adjusted personalized fusion parameters corresponding to the key target.

9. A resource recommendation device, comprising: An acquisition module is used to obtain the multi-objective prediction results of each candidate resource, the comprehensive user characteristics of the target user, and the key user characteristics; A first feature processing module is used to extract the key user features using the first neural network in the target evolution strategy model to obtain a first feature vector; a second feature processing module, configured to process the comprehensive user features and the first feature vector using a second neural network in the target evolution strategy model to obtain personalized fusion parameters for each target corresponding to the target user; a determination module, configured to determine a resource score of the candidate resource based on the personalized fusion parameter and the multi-objective prediction result; The recommendation module is used to recommend resources to the target user based on the resource scores of the candidate resources.

10. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-8.

12. A computer program product comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Images