Resource recall model updating method, device, electronic device and storage medium
By forming a click resource pool, selecting balanced positive and negative samples, and optimizing model parameters, the resource recall model in media platforms addresses the accuracy issues of existing models, enhancing the relevance of resource delivery.
Patent Information
- Application Number
- CN202011288307.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-11-17
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2040-11-17
AI Technical Summary
The existing advertising recall models in media platforms suffer from poor accuracy due to an unbalanced ratio of positive and negative samples in training data, leading to insufficient learning of useful information.
A method to update the resource recall model by forming a click resource pool, selecting positive and negative samples, generating training samples with user characteristics, and optimizing model parameters using a preset loss function to balance sample ratios and improve learning.
The updated model accurately learns useful information, enhancing the accuracy of resource recall by balancing positive and negative samples, thus improving the relevance of resource delivery to users.
Smart Images

Figure CN114519593B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer technology, and in particular to a resource recall model updating method, device, electronic device and storage medium. Background Art
[0002] At present, the business processing flow of the advertising system of each media platform generally includes the targeting stage, the recall stage, the sorting stage and the display stage. In the targeting stage, the advertising system receives the targeting information set by the advertiser for the advertising, such as crowd targeting, user information and regional targeting, etc. This targeting information can help advertisers better circle their target users; in the recall stage, the advertising system preliminarily screens the advertisements in the advertising library according to the advertising push request; in the sorting stage, the advertising system sorts the preliminarily screened advertisements, and further screens the final advertisements for delivery based on the sorting results; in the display stage, the advertising system sends the screening results of the sorting stage to the user client for advertising exposure. From the business processing flow of the above advertising system, it can be seen that the recall stage serves the subsequent sorting stage. The accuracy of the recall directly affects the accuracy of the subsequent sorted candidate set, which in turn has a crucial impact on the accuracy of the advertisements finally pushed to the user.
[0003] In the related art, the advertising system usually uses a trained advertising recall model to recall advertisements during the recall stage. The advertising recall model generally adopts online streaming training. The positive samples in the training data used are advertisements that have been clicked by users, and the negative samples are advertisements that have been shown to users but not clicked. Usually, the number of negative samples far exceeds the number of positive samples, and the ratio of positive and negative samples is unbalanced. The feedback of positive samples is easily submerged in the feedback of negative samples, resulting in the advertising recall model being unable to fully learn the useful information in the training data, which in turn leads to poor accuracy of advertising recall. Summary of the invention
[0004] The present disclosure provides a resource recall model update method, device, electronic device and storage medium to at least solve the problem in the related art that the resource recall model for resources such as advertisements cannot fully learn the useful information in the training data, resulting in poor accuracy of resource recall. The technical solution of the present disclosure is as follows:
[0005] According to a first aspect of an embodiment of the present disclosure, a resource recall model updating method is provided, comprising:
[0006] Get the resource click log corresponding to the current model update cycle;
[0007] Determine a click resource pool according to the resource click log, the click resource pool including at least one click resource, the click resource representing a resource clicked by a user within a first historical time period, the first historical time period being a time period between a current model update cycle and a previous model update cycle;
[0008] A first click resource in the click resource pool is selected, the first click resource is used as a positive sample resource, and a negative sample resource corresponding to the positive sample resource is determined from the remaining click resources; the first click resource is any click resource in the click resource pool, and the remaining click resources are click resources in the click resource pool other than the first click resource;
[0009] Generate a training sample in a training sample set according to a positive sample resource, a negative sample resource corresponding to the positive sample resource, and a user who clicks the positive sample resource;
[0010] According to the training sample set, the parameters in the resource recall model are optimized using a preset loss function to obtain an updated resource recall model; the preset loss function is determined based on the matching degree between the user and the positive sample resources in the training sample and the matching degree between the user and the negative sample resources.
[0011] In an exemplary implementation, generating a training sample in a training sample set according to a positive sample resource, a negative sample resource corresponding to the positive sample resource, and a user who clicks the positive sample resource includes:
[0012] Determine a user who clicks the positive sample resource, and obtain user characteristics of the user;
[0013] Determining positive sample resource characteristics of the positive sample resource;
[0014] Determining negative sample resource characteristics of negative sample resources corresponding to the positive sample resources;
[0015] The user features, the positive sample resource features and the negative sample resource features constitute training samples in a training sample set.
[0016] In an exemplary implementation, the optimizing the parameters in the resource recall model using a preset loss function according to the training sample set includes:
[0017] For each training sample in the training sample set, inputting the user features, positive sample resource features and negative sample resource features included in the training sample into the resource recall model;
[0018] Obtaining a user feature vector according to the user features through the user neural network of the resource recall model;
[0019] Obtaining a positive sample resource vector and a negative sample resource vector respectively according to the positive sample resource features and the negative sample resource features through the resource neural network of the resource recall model;
[0020] Determining a first matching degree between the user feature vector and the positive sample resource vector, and determining a second matching degree between the user feature vector and the negative sample resource vector;
[0021] The preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree.
[0022] In an exemplary embodiment, the preset loss function includes a first difference term, a second difference term and a maximum value term;
[0023] The preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree, including:
[0024] Determine a first difference between the second matching degree and the first matching degree through the first difference item;
[0025] Determine a second difference between the first difference and a preset fixed value by using the second difference item;
[0026] Taking the maximum value between the second difference and the value zero through the maximum value item, and taking the maximum value as the function value of the preset loss function;
[0027] The parameters in the resource recall model are adjusted in a direction to minimize the function value until a training end condition is met.
[0028] In an exemplary embodiment, the method further comprises:
[0029] In response to a resource push request from a target user, determining a candidate resource set matching the target user according to the directional information of each resource in the resource library;
[0030] Determining target user characteristics of the target user and candidate resource characteristics of each candidate resource in the candidate resource set;
[0031] Inputting the target user characteristics and the candidate resource characteristics of each candidate resource into the updated resource recall model to obtain the matching degree between the target user and each candidate resource;
[0032] Determining at least one target candidate resource from the candidate resource set according to the matching degree between the target user and each candidate resource;
[0033] A return result corresponding to the resource push request is determined according to the at least one target candidate resource, where the return result includes a target resource to be displayed.
[0034] In an exemplary implementation, after determining the return result corresponding to the resource push request according to the at least one target candidate resource, the method further includes:
[0035] In response to a click operation of the target user on the target resource, acquiring a user identifier of the target user and a resource identifier of the target resource;
[0036] A resource click log corresponding to the next model update cycle is generated according to the user identifier of the target user and the resource identifier of the target resource.
[0037] According to a second aspect of an embodiment of the present disclosure, a resource recall model updating device is provided, comprising:
[0038] A click log acquisition unit is configured to execute acquisition of resource click logs corresponding to the current model update cycle;
[0039] A first determining unit is configured to determine a click resource pool according to the resource click log, wherein the click resource pool includes at least one click resource, and the click resource represents a resource clicked by a user in a first historical time period, wherein the first historical time period is a time period between a current model update cycle and a previous model update cycle;
[0040] a negative sampling unit, configured to select a first click resource in the click resource pool, take the first click resource as a positive sample resource, and determine a negative sample resource corresponding to the positive sample resource from the remaining click resources; the first click resource is any click resource in the click resource pool, and the remaining click resources are click resources in the click resource pool other than the first click resource;
[0041] A training data generating unit is configured to generate training samples in a training sample set according to positive sample resources, negative sample resources corresponding to the positive sample resources, and users who click on the positive sample resources;
[0042] A model updating unit is configured to optimize the parameters in the resource recall model using a preset loss function according to the training sample set to obtain an updated resource recall model; the preset loss function is determined based on the degree of matching between the user and the positive sample resources in the training sample and the degree of matching between the user and the negative sample resources.
[0043] In an exemplary embodiment, the training data generating unit includes:
[0044] A first determining unit is configured to determine a user who clicks the positive sample resource and obtain a user feature of the user;
[0045] A second determining unit is configured to determine a positive sample resource feature of the positive sample resource;
[0046] A third determining unit is configured to determine a negative sample resource feature of a negative sample resource corresponding to the positive sample resource;
[0047] The generating subunit is configured to execute the formation of training samples in a training sample set by using the user features, the positive sample resource features and the negative sample resource features.
[0048] In an exemplary embodiment, the model updating unit includes:
[0049] An input unit is configured to execute, for each training sample in the training sample set, inputting the user features, positive sample resource features, and negative sample resource features included in the training sample into the resource recall model;
[0050] A first network unit is configured to execute a user neural network through the resource recall model and obtain a user feature vector according to the user feature;
[0051] A second network unit is configured to execute a resource neural network through the resource recall model, and obtain a positive sample resource vector and a negative sample resource vector according to the positive sample resource feature and the negative sample resource feature respectively;
[0052] A matching unit, configured to determine a first matching degree between the user feature vector and the positive sample resource vector, and determine a second matching degree between the user feature vector and the negative sample resource vector;
[0053] The optimization unit is configured to execute a preset loss function to optimize the parameters in the resource recall model based on the first matching degree and the second matching degree.
[0054] In an exemplary embodiment, the preset loss function includes a first difference term, a second difference term and a maximum value term; and the optimization unit includes:
[0055] A fourth determining unit is configured to determine a first difference between the second matching degree and the first matching degree by using the first difference item;
[0056] A fifth determining unit is configured to determine a second difference between the first difference and a preset fixed value by using the second difference item;
[0057] A sixth determining unit is configured to perform the step of taking the maximum value between the second difference and the value zero through the maximum value item, and taking the maximum value as the function value of the preset loss function;
[0058] The parameter adjustment unit is configured to adjust the parameters in the resource recall model in a direction of minimizing the function value until a training end condition is met.
[0059] In an exemplary embodiment, the device further comprises:
[0060] A request response unit, configured to execute a response to a resource push request of a target user, and determine a candidate resource set matching the target user according to the directional information of each resource in the resource library;
[0061] a seventh determining unit, configured to determine a target user feature of the target user and a candidate resource feature of each candidate resource in the candidate resource set;
[0062] a matching degree prediction unit configured to input the target user characteristics and the candidate resource characteristics of each candidate resource into the updated resource recall model to obtain the matching degree between the target user and each candidate resource;
[0063] An eighth determining unit is configured to determine at least one target candidate resource from the candidate resource set according to a matching degree between the target user and each candidate resource;
[0064] The ninth determining unit is configured to determine a return result corresponding to the resource push request according to the at least one target candidate resource, wherein the return result includes a target resource to be displayed.
[0065] In an exemplary embodiment, the device further comprises:
[0066] an acquisition unit, configured to acquire a user identifier of the target user and a resource identifier of the target resource in response to a click operation of the target user on the target resource;
[0067] The click log generating unit is configured to generate a resource click log corresponding to the next model update cycle according to the user identifier of the target user and the resource identifier of the target resource.
[0068] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:
[0069] processor;
[0070] a memory for storing instructions executable by the processor;
[0071] The processor is configured to execute the instructions to implement the resource recall model updating method as described in any of the above implementations.
[0072] According to a fourth aspect of an embodiment of the present disclosure, a storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the resource recall model updating method described in any of the above embodiments.
[0073] According to a fifth aspect of the embodiments of the present disclosure, a computer program product is provided, the computer program product comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the resource recall model updating method provided in any one of the above-mentioned embodiments;
[0074] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:
[0075] A click resource pool is formed by all resources clicked by users in the time period between the current model update cycle and the previous model update cycle, and negative sampling is performed on the click resource pool. Any click resource in the click resource pool is taken as a positive sample resource, and a negative sample resource corresponding to the positive sample resource is determined from the remaining click resources. The training samples in the training sample set are generated in combination with the users who click on the positive sample resource, so that the ratio of positive and negative samples used for training is balanced. Then, according to the training sample set, the preset loss function is used to optimize the parameters in the resource recall model to obtain an updated resource recall model. The resource recall model can fully learn the useful information in the training data that is more matched with the resource recall process, thereby improving the accuracy of resource recall based on the resource recall model.
[0076] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0077] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.
[0078] Figure 1 is an application environment diagram of a resource recall model updating method according to an exemplary embodiment;
[0079] Figure 2 is a flow chart of a resource recall model updating method according to an exemplary embodiment;
[0080] Figure 3 is a schematic diagram showing updating of resource recall model training according to an exemplary embodiment;
[0081] Figure 4 is a flow chart of another resource recall model updating method according to an exemplary embodiment;
[0082] Figure 5 is a block diagram of a resource recall model updating device according to an exemplary embodiment;
[0083] Figure 6 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION
[0084] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.
[0085] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0086] See also Figure 1 , which shows an application environment diagram of a resource recall model update method according to an exemplary embodiment. The application environment may include a server 110 and multiple terminals 120. The server 110 and the multiple terminals 120 may be connected via a wired network or a wireless network.
[0087] The multiple terminals 120 may be smart phones, tablet computers, laptop computers, desktop computers, etc., but are not limited thereto. The terminal 120 may be installed with client software such as an application (Application, referred to as App) that provides corresponding business functions. The business functions may include, but are not limited to, a live broadcast function and a short video playback function, that is, the application may include, but are not limited to, a live broadcast App and a short video App. The user of the terminal 120 may log in to the application using pre-registered user information, which may include an account number and a password.
[0088] The server 110 may be a server that provides background services for the application in the terminal 120, or may be another server that is connected and communicated with the background server of the application, and may be a single server or a server cluster composed of multiple servers. The server 110 may provide a resource push service, for example, when a user logs in to the application or refreshes a page of the application, the server 110 may push resources such as advertisements to the user.
[0089] In an application scenario, the server 110 may be a server for providing an advertisement recall service in an advertisement system. When providing the advertisement recall service, the server 110 may recall advertisements in the advertisement system based on an advertisement recall model, and the advertisement recall model may be updated using the online streaming training method of the embodiment of the present disclosure. The model update cycle may be set during the online streaming training update process, and the server 110 may train and update the advertisement recall model when the model update cycle is reached. For example, the server 110 may train and update the advertisement recall model once every 15 minutes. Taking the time period from 9:30 to 10:00 as an example, the server 110 performs a training update on the advertisement recall model at 9:30, and the next model update cycle is 9:45. When the server 110 reaches 9:45, it may train and update the advertisement recall model updated at 9:30 in the previous model update cycle. Similarly, when the server 110 reaches 10:00, it may train and update the advertisement recall model updated at 9:45 in the previous model update cycle.
[0090] It is understandable that the resources in the embodiments of the present disclosure are not limited to advertisements, but may also include other resources that can be pushed to users of the application, such as news, videos, etc., which are not specifically limited in the embodiments of the present disclosure.
[0091] The resource recall model updating method of the embodiment of the present disclosure is described in detail below by taking a model updating cycle, namely the current model updating cycle, as an example.
[0092] Figure 2 is a flow chart of a resource recall model method according to an exemplary embodiment. Figure 2 As shown, the resource recall model update method is applied to Figure 1 The server 110 shown is illustrated and includes the following steps.
[0093] In step S210, the resource click log corresponding to the current model update cycle is obtained.
[0094] In actual applications, after the server pushes the resources to the user of the application and displays them through the corresponding terminal, the user can choose to click on the displayed resource or not according to his or her interest in the displayed resource. When the user clicks on the displayed resource, the server can obtain the user ID of the user and the resource ID of the clicked resource in response to the user's click operation on the resource, and generate a resource click log based on the user ID and the resource ID of the clicked resource.
[0095] The server can store the generated resource click logs. Since the server trains and updates the resource recall model according to the set model update cycle in the embodiment of the present disclosure, the server can store the resource click logs in time periods, and store the resource click logs received in the time period between two adjacent model update cycles as a batch. For example, if the time period between two adjacent model update cycles is 9:30 to 9:45, the server will store the resource click logs received between 9:30 and 9:45 as a batch. When the current model update cycle, such as 9:45, is reached, the server can obtain the resource click log corresponding to the current model update cycle, which is the resource click log in the time period between the current model update cycle and the previous model update cycle, such as 9:30 to 9:45.
[0096] In a specific implementation, the server can set an update timer, and through the timing function of the update timer, trigger the server to train and update the resource recall model when the current model update cycle is reached. For example, the update timer can be set to trigger the server to execute the above-mentioned action of obtaining the resource click log corresponding to the current model update cycle every 15 minutes, and then start training and updating the resource recall model.
[0097] In step S220, a click resource pool is determined according to the resource click log.
[0098] The click resource pool includes at least one click resource, which represents a resource clicked by a user within a first historical time period, and the first historical time period is a time period between a current model update cycle and a previous model update cycle.
[0099] Specifically, the server can obtain the resource identifier in the resource click log, search for resources matching the resource identifier in the resource library, and put the resources matching the resource identifier in the resource library into the click resource pool as click resources. The resource library stores all resources to be pushed, and the resource library can be located locally on the server or in a distributed database system.
[0100] In step S230, the first click resource in the click resource pool is selected, the first click resource is taken as the positive sample resource, and the negative sample resource corresponding to the positive sample resource is determined from the remaining click resources.
[0101] Among them, the first click resource is any click resource in the above-mentioned click resource pool, and the remaining click resources are click resources in the above-mentioned click resource pool except the first click resource.
[0102] The click resource pool includes (i1, i2, ..., i N ), where i represents the above-mentioned click resource, then for any click resource i k (1≤k≤N), the server takes the click resource as a positive sample resource From the remaining click resources (i1, ..., i k-1 ,i k+1 …,i N ) to determine the positive sample resource Corresponding negative sample resources Specifically, we can select the remaining click resources (i1, i k-1 ,i k+1 …,i N ) is randomly selected as a positive sample resource Corresponding negative sample resources Thus, we get the positive sample resource-negative sample resource pair, namely in Then based on the above sampling process, for the click resource pool (i1, i2, ..., i N ) can get the following results: It can be seen that the ratio of positive and negative sample resources obtained after sampling processing is balanced, and both positive and negative sample resources are resources clicked by users during the model update cycle, which can avoid the feedback of positive samples being overwhelmed by the feedback of negative samples in subsequent training.
[0103] In step S240, a training sample in a training sample set is generated according to the positive sample resource, the negative sample resource corresponding to the positive sample resource, and the user who clicks the positive sample resource.
[0104] In a possible implementation, the server may generate training samples in the training sample set according to the user, the positive sample resource clicked by the user, and the negative sample resource corresponding to the positive sample resource, including the following steps:
[0105] (1) Determine the user who clicks the positive sample resource and obtain the user characteristics of the user.
[0106] Specifically, the server can obtain the user ID corresponding to each click resource from the click resource log. The user ID is used to identify the user who clicked the click resource. The user portrait is obtained through the user ID. The user portrait is used to outline the user at the data level. Specifically, it can be composed of multiple user tags. In the era of big data, user information is flooded in the network. In order to facilitate the processing of user information, various specific information of the user can be abstracted into tags, and these tags are used to concretize the user image. The user tag can include but is not limited to age, gender, region, interest resource category, membership, etc. The server can directly determine the user portrait corresponding to the user who clicked the positive sample resource as the user feature of the user, and can also determine one or more user tags in the user portrait as the user feature of the user.
[0107] (2) Determine the positive sample resource characteristics of the positive sample resources.
[0108] (3) Determine the negative sample resource characteristics of the negative sample resources corresponding to the above positive sample resources.
[0109] In the embodiments of the present disclosure, the resource features of each resource may include but are not limited to resource category, resource orientation information, resource price, resource push record, etc. The server may store and update the resource features of each resource in the resource library in real time. The server may determine one or more of the resource features corresponding to the positive sample resource (such as resource category, resource orientation information, resource price, resource push record, etc.) as the positive sample resource feature of the positive sample resource; similarly, the server may determine one or more of the resource features corresponding to the negative sample resource (such as resource category, resource push condition, resource price, resource push record, etc.) as the negative sample resource feature.
[0110] (4) The above user features, positive sample resource features and corresponding negative sample resource features constitute the training samples in the training sample set.
[0111] Specifically, with u k Indicates a positive sample resource for clicks The user, F(u k ) represents user u k The user characteristics of the user u k Clicked positive sample resources The positive sample resource feature is expressed as The positive sample resource Corresponding negative sample resources The negative sample resource feature is expressed as The server can form a training sample The training sample is a triple. Then, for the click resource pool (i1, i2, ..., i N) The training sample set generated by can be expressed as:
[0112] The embodiments of the present disclosure construct, for each positive sample resource, a training sample of a triplet structure by obtaining the user characteristics of the user who clicks the positive sample resource, the positive sample resource characteristics of the positive sample resource, and the negative sample resource characteristics of the negative sample resource corresponding to the positive sample resource, so as to fully learn the useful information in the training sample set during subsequent model training, thereby adjusting the model parameters more accurately.
[0113] In step S250, the parameters in the resource recall model are optimized using a preset loss function according to the training sample set to obtain an updated resource recall model.
[0114] The preset loss function is determined based on the matching degree between the user and the positive sample resources in the above training samples and the matching degree between the user and the negative sample resources. The preset loss function can optimize the order between the positive sample resources and the negative sample resources, that is, the updated resource recall model obtained by optimizing the parameters in the resource recall model using the preset loss function according to the above training sample set can learn the order relationship in the training samples, and the resource recall also involves an order relationship, so that the updated resource recall model is more matched with the actual process of resource recall, which is conducive to the resource recall model to fully learn the useful information in the training data that is more matched with the actual recall process.
[0115] In the embodiments of the present disclosure, Figure 3 As shown, the resource recall model may include a user neural network, a resource neural network and a matching node, wherein the user neural network is used to vectorize user features, and the resource neural network is used to vectorize positive sample resource features and negative sample resource features, respectively, and the positive sample resource features and the negative sample resource features share a resource neural network. The output ends of the user neural network and the resource neural network are respectively connected to the matching node, and the matching node is used to calculate the matching degree between the user and the positive sample resource and the matching degree between the user and the negative sample resource.
[0116] Among them, the user neural network and the resource neural network can map the input features to the same k-dimensional embedding space. Exemplarily, both the user neural network and the resource neural network can be DNN (Deep Neural Network) networks, and the DNN network includes an input layer, a hidden layer, and an output layer. The layers are fully connected, that is, any neuron in the i-th layer is connected to any neuron in the i+1-th layer.
[0117] Based on this, the server may include the following steps when optimizing the parameters in the resource recall model using a preset loss function according to the training sample set:
[0118] (1) For each training sample in the training sample set, the user features, positive sample resource features, and negative sample resource features included in the training sample are input into the resource recall model.
[0119] (2) The user neural network of the resource recall model is used to obtain a user feature vector based on the above user features.
[0120] (3) Through the resource neural network of the resource recall model, the positive sample resource vector and the negative sample resource vector are obtained according to the above positive sample resource characteristics and negative sample resource characteristics.
[0121] (4) Determine a first matching degree between the user feature vector and the positive sample resource vector through a matching neural network of a resource recall model, and determine a second matching degree between the user feature vector and the negative sample resource vector.
[0122] With training samples For example, please continue to see Figure 3 After receiving the input training sample, the resource recall model converts the user feature F(u k ) is input into the user neural network, and the positive sample resource features in the training sample are and negative sample resource features Input into the resource neural network, and obtain the user feature F(u k ) is the vector expression of the user feature vector q k , and obtain the positive sample resource features through the resource neural network The vector expression of the positive sample resource vector and negative sample resource features The vector expression of negative sample resource vector The above vector expression process can be expressed as:
[0123]
[0124]
[0125]
[0126] Among them, FC represents the fully connected layer, θ1 represents the parameters to be optimized in the user neural network, and θ2 represents the parameters to be optimized in the resource neural network.
[0127] User feature vector q k , positive sample resource vector and negative sample resource vector As the input of the matching node, the matching node calculates the user feature vector q k and the positive sample resource vector The first matching degree between and the user feature vector q k and negative sample resource vector The second matching degree between them, for example, can be characterized by cosine similarity.
[0128] (5) The preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree.
[0129] In the embodiment of the present disclosure, the parameters to be optimized in the resource recall model include the parameters θ1 to be optimized in the user neural network and the parameters θ2 to be optimized in the resource neural network. Through the vector expression of user features by the user neural network, and the vector expression of positive sample resource features and negative sample resource features by the resource neural network, and based on the results of the vector expression, a first matching degree is obtained to characterize the matching degree between the user and the positive sample resource, and a second matching degree is obtained to characterize the matching degree between the user and the negative sample resource, so that the preset loss function can optimize the parameters of the user neural network and the resource neural network in the resource recall model based on the first matching degree and the second matching degree to obtain an updated resource recall model.
[0130] Specifically, the preset loss function may include a first difference term, a second difference term, and a maximum value term. When the preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree, the first difference between the second matching degree and the first matching degree may be calculated through the first difference term, the second difference between the first difference and the preset fixed value may be calculated through the second difference term, and the maximum value between the second difference and the value zero may be taken through the maximum value term, and the maximum value may be used as the function value of the preset loss function, and then the parameters in the resource recall model may be adjusted in the direction of minimizing the function value until the training end condition is met. Exemplarily, the preset loss function may be expressed as:
[0131]
[0132] Wherein, θ represents the parameters of the resource recall model, including θ1 and θ2 mentioned above in the embodiment of the present disclosure; Indicates the first matching degree; Indicates the second matching degree; represents the first difference; Represents the second difference; m represents a preset fixed value, and the specific value of the preset fixed value can be set according to actual needs. For example, m can be set in the setting range of 0.2 to 0.3. The preset loss function Loss controls the error between the positive sample resource and the negative sample resource between 0 and m. The preset loss function constructed by the above-mentioned first difference term, second difference term and maximum value term realizes the optimization of the parameters in the resource recall model based on the first matching degree and the second matching degree, so that the resource recall model can more fully learn the order relationship between the sample resources in the training sample set during the update process, thereby making the updated resource recall model have better resource recall accuracy.
[0133] Exemplarily, the gradient descent method may be used to adjust the parameters in the resource recall model in the direction of minimizing the function value. The training end condition may be, but is not limited to, the number of iterations reaching a preset number threshold, for example, the preset number threshold may be 100 times.
[0134] The embodiments of the present disclosure form a click resource pool through all resources clicked by users in the time period between the current model update cycle and the previous model update cycle, perform negative sampling on the click resource pool, take any click resource in the click resource pool as a positive sample resource, determine the negative sample resource corresponding to the positive sample resource from the remaining click resources, and generate training samples in the training sample set in combination with the user who clicks the positive sample resource, so that the ratio of positive and negative samples used for training is balanced, and then optimize the parameters in the resource recall model according to the training sample set using a preset loss function, thereby avoiding the feedback of positive samples being overwhelmed by the feedback of negative samples during the training process, improving the training ability of the resource recall model, and the resource recall model can fully learn the useful information in the training data that is more matched with the resource recall process, thereby improving the accuracy of the updated resource recall model for resource recall.
[0135] Figure 4 is a flowchart of another resource recall model updating method according to an exemplary embodiment. Figure 4 As shown, after optimizing the parameters in the resource recall model using a preset loss function according to the training sample set in step S250 to obtain an updated resource recall model, the method may further include:
[0136] In step S410, in response to a resource push request from a target user, a candidate resource set matching the target user is determined according to the directional information of each resource in the resource library.
[0137] Among them, the targeting information refers to the resource push conditions corresponding to the resource, that is, a resource is only used as a resource to be recalled when it meets the set resource push conditions. Taking the resource as an advertisement as an example, the targeting information of the advertisement may include user age, user gender, regional information, etc. The targeting information of the resource can be set by the resource publisher. After obtaining the targeting information set by the resource publisher, the resource push system can store the targeting information of each resource accordingly. The target user can be any user who triggers the resource push request.
[0138] Specifically, when the target user logs in to the application or refreshes the page of the application, the resource push instruction can be triggered to send a resource push request to the server. Correspondingly, the server receives the resource push request, and in response to the resource push request, searches for resources that match the target user according to the directional information of each resource in the resource library, and each resource that matches the target user constitutes a candidate resource set.
[0139] In step S420, target user characteristics of the target user and candidate resource characteristics of each candidate resource in the candidate resource set are determined.
[0140] Specifically, the resource push request may carry the user ID of the target user, and the server may obtain the target user portrait of the target user based on the user ID of the target user, and determine the target user characteristics of the target user based on one or more user tags included in the target user portrait, where the user tags may include but are not limited to age, gender, region, category of resource of interest, membership, etc. For example, the server may directly determine the target user portrait as the target user characteristics of the target user, or may determine one or more user tags in the target user portrait as the target user characteristics of the target user.
[0141] The server may determine corresponding candidate resource features based on resource features of each candidate resource in the candidate resource set. The candidate resource features may include but are not limited to resource category, resource orientation information, resource price, resource push record, and the like.
[0142] In step S430, the target user characteristics and the candidate resource characteristics of each candidate resource are input into the updated resource recall model to obtain the matching degree between the target user and each candidate resource.
[0143] Among them, the matching degree represents the degree of matching between the target user and the candidate resources. The matching degree may include but is not limited to cosine similarity. The matching degree may represent the probability of the target user clicking on the corresponding candidate resource. Generally, the larger the matching degree, the greater the probability of the target user clicking on the corresponding candidate resource. Conversely, the smaller the matching degree, the smaller the probability of the target user clicking on the corresponding candidate resource.
[0144] In step S440, at least one target candidate resource is determined from the candidate resource set according to the matching degree between the target user and each candidate resource.
[0145] Specifically, the candidate resources in the candidate resource set may be arranged in descending order according to the matching degree, and the top N candidate resources may be recalled as target candidate resources.
[0146] In step S450, a return result corresponding to the resource push request is determined according to the at least one target candidate resource, and the return result includes the target resource to be displayed.
[0147] Specifically, in step S450, the recalled target candidate resources can be further screened in combination with other screening conditions to obtain the target resources for final push, and the other screening conditions can be set according to the needs of the actual application. Taking the resource as an advertisement as an example, the other screening condition can be the product of the matching degree of the target candidate resource and the customer's bid for the advertisement, and the target candidate resource with the largest product result is returned to the target user as the target resource to be displayed.
[0148] Since the updated resource recall model has fully learned the useful information in the training data that is more consistent with the resource recall process in the previous model update cycle, it improves the accuracy of subsequent resource recalls and thus improves the accuracy of resources pushed to users.
[0149] In a specific embodiment, Figure 4 As shown, after the server determines the return result corresponding to the resource push request according to the at least one target candidate resource, the server may further include:
[0150] S460 , in response to a click operation of the target user on the target resource, obtaining a user identifier of the target user and a resource identifier of the target resource.
[0151] S470: Generate a resource click log corresponding to the next model update cycle according to the user identifier of the target user and the resource identifier of the target resource.
[0152] Specifically, after the server pushes the target resource to the target user and displays it through the corresponding terminal, the target user can choose to click on the target resource or not according to his or her interest in the target resource. After the target user clicks on the target resource, the server can obtain the user ID of the target user and the resource ID of the target resource in response to the target user's click operation on the target resource, and generate a resource click log corresponding to the next model update cycle based on the user ID of the target user and the resource ID of the target resource. When the next model update cycle arrives, the server can repeat the aforementioned steps S210 to S250 in the embodiment of the present disclosure according to the resource click log generated above to update the resource recall model again.
[0153] Figure 5 is a block diagram of a resource recall model updating device according to an exemplary embodiment. Figure 5 The device includes a click log obtaining unit 510, a first determining unit 520, a negative sampling unit 530, a training data generating unit 540 and a model updating unit 550.
[0154] The click log acquisition unit 510 is configured to acquire the resource click log corresponding to the current model update cycle;
[0155] The first determining unit 520 is configured to determine a click resource pool according to the resource click log, where the click resource pool includes at least one click resource, where the click resource represents a resource clicked by a user in a first historical time period, where the first historical time period is a time period between a current model update cycle and a previous model update cycle;
[0156] The negative sampling unit 530 is configured to select a first click resource in the click resource pool, take the first click resource as a positive sample resource, and determine a negative sample resource corresponding to the positive sample resource from the remaining click resources; the first click resource is any click resource in the click resource pool, and the remaining click resources are click resources in the click resource pool other than the first click resource;
[0157] The training data generating unit 540 is configured to generate a training sample in a training sample set according to a positive sample resource, a negative sample resource corresponding to the positive sample resource, and a user who clicks the positive sample resource;
[0158] The model updating unit 550 is configured to optimize the parameters in the resource recall model using a preset loss function according to the training sample set to obtain an updated resource recall model; the preset loss function is determined based on the degree of matching between the user and the positive sample resources in the training sample and the degree of matching between the user and the negative sample resources.
[0159] In an exemplary embodiment, the training data generating unit 540 may include:
[0160] A first determining unit is configured to determine a user who clicks a positive sample resource and obtain a user feature of the user;
[0161] A second determining unit is configured to determine the positive sample resource features of the positive sample resource;
[0162] A third determining unit is configured to determine negative sample resource features of negative sample resources corresponding to the positive sample resources;
[0163] The generating subunit is configured to execute the training samples in the training sample set formed by the user features, the positive sample resource features and the negative sample resource features.
[0164] In an exemplary implementation, the model updating unit 550 may include:
[0165] An input unit is configured to execute, for each training sample in the training sample set, inputting the user features, positive sample resource features, and negative sample resource features included in the training sample into the resource recall model;
[0166] A first network unit is configured to execute a user neural network through a resource recall model to obtain a user feature vector according to user features;
[0167] The second network unit is configured to execute a resource neural network through a resource recall model, and obtain a positive sample resource vector and a negative sample resource vector according to the positive sample resource features and the negative sample resource features respectively;
[0168] A matching unit, configured to determine a first matching degree between the user feature vector and the positive sample resource vector, and determine a second matching degree between the user feature vector and the negative sample resource vector;
[0169] The optimization unit is configured to execute a preset loss function to optimize the parameters in the resource recall model based on the first matching degree and the second matching degree.
[0170] In an exemplary embodiment, the preset loss function includes a first difference term, a second difference term and a maximum value term; the above-mentioned optimization unit may include:
[0171] A fourth determining unit is configured to determine a first difference between the second matching degree and the first matching degree through the first difference item;
[0172] A fifth determining unit is configured to determine a second difference between the first difference and a preset fixed value through the second difference item;
[0173] A sixth determining unit is configured to perform the maximum value between the second difference and the value zero through the maximum value item, and use the maximum value as the function value of the preset loss function;
[0174] The parameter adjustment unit is configured to adjust the parameters in the resource recall model in a direction of minimizing the above function value until the training end condition is met.
[0175] In an exemplary embodiment, the device may further include:
[0176] A request response unit, configured to execute a response to a resource push request of a target user, and determine a candidate resource set matching the target user according to the directional information of each resource in the resource library;
[0177] a seventh determining unit, configured to determine a target user feature of the target user and a candidate resource feature of each candidate resource in the candidate resource set;
[0178] A matching degree prediction unit is configured to input the target user characteristics and the candidate resource characteristics of each candidate resource into the updated resource recall model to obtain the matching degree between the target user and each candidate resource;
[0179] An eighth determination unit is configured to determine at least one target candidate resource from the candidate resource set according to a matching degree between the target user and each candidate resource;
[0180] The ninth determining unit is configured to determine a return result corresponding to the resource push request according to the at least one target candidate resource, where the return result includes a target resource to be displayed.
[0181] In an exemplary embodiment, the device may further include:
[0182] an acquisition unit, configured to execute, in response to a click operation of a target user on a target resource, acquire a user identifier of the target user and a resource identifier of the target resource;
[0183] The click log generating unit is configured to generate a resource click log corresponding to the next model update cycle according to the user identifier of the target user and the resource identifier of the target resource.
[0184] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0185] In an exemplary embodiment, an electronic device is also provided, including a processor; a memory for storing processor executable instructions; wherein the processor is configured to implement the steps of any resource recall model update method in the above embodiments when executing the instructions stored in the memory.
[0186] The electronic device may be a terminal, a server or a similar computing device. For example, the electronic device is a server. Figure 6 It is a block diagram of an electronic device for updating a resource recall model according to an exemplary embodiment. The electronic device 600 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 610 (the processor 610 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 630 for storing data, and one or more storage media 620 (such as one or more mass storage devices) for storing application programs 623 or data 622. Among them, the memory 630 and the storage medium 620 can be short-term storage or permanent storage. The program stored in the storage medium 620 may include one or more modules, and each module may include a series of instruction operations in the electronic device. Furthermore, the central processor 610 may be configured to communicate with the storage medium 620 to execute a series of instruction operations in the storage medium 620 on the electronic device 600. The electronic device 600 may also include one or more power supplies 660, one or more wired or wireless network interfaces 650, one or more input and output interfaces 640, and / or one or more operating systems 621, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.
[0187] The input / output interface 640 may be used to receive or send data via a network. The specific example of the network may include a wireless network provided by a communication provider of the electronic device 600. In one example, the input / output interface 640 includes a network adapter (Network Interface Controller, NIC), which may be connected to other network devices via a base station so as to communicate with the Internet. In an exemplary embodiment, the input / output interface 640 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0188] It can be understood by those skilled in the art that Figure 6 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 6 More or fewer components as shown, or with Figure 6 Different configurations shown.
[0189] In an exemplary embodiment, a storage medium is further provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the steps of any resource recall model updating method in the above embodiments.
[0190] In an exemplary embodiment, a computer program product is also provided, the computer program product includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the resource recall model update method provided in any of the above embodiments.
[0191] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0192] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0193] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A resource recall model updating method, characterized in that: include: Get the resource click log corresponding to the current model update cycle; Determine a click resource pool according to the resource click log, the click resource pool including at least one click resource, the click resource representing a resource clicked by a user within a first historical time period, the first historical time period being a time period between a current model update cycle and a previous model update cycle; A first click resource in the click resource pool is selected, the first click resource is used as a positive sample resource, and a negative sample resource corresponding to the positive sample resource is determined from the remaining click resources; the first click resource is any click resource in the click resource pool, and the remaining click resources are click resources in the click resource pool other than the first click resource; Generate a training sample in a training sample set according to a positive sample resource, a negative sample resource corresponding to the positive sample resource, and a user who clicks the positive sample resource; According to the training sample set, the parameters in the resource recall model are optimized using a preset loss function to obtain an updated resource recall model; the preset loss function is determined based on the matching degree between the user and the positive sample resources in the training sample and the matching degree between the user and the negative sample resources.
2. The resource recall model updating method according to claim 1, characterized in that: The step of generating training samples in a training sample set according to positive sample resources, negative sample resources corresponding to the positive sample resources, and users who click on the positive sample resources comprises: Determine a user who clicks the positive sample resource, and obtain user characteristics of the user; Determining positive sample resource characteristics of the positive sample resource; Determining negative sample resource characteristics of negative sample resources corresponding to the positive sample resources; The user features, the positive sample resource features and the negative sample resource features constitute training samples in a training sample set.
3. The resource recall model updating method according to claim 2, characterized in that: The optimizing the parameters in the resource recall model using a preset loss function according to the training sample set includes: For each training sample in the training sample set, inputting the user features, positive sample resource features and negative sample resource features included in the training sample into the resource recall model; Obtaining a user feature vector according to the user features through the user neural network of the resource recall model; Obtaining a positive sample resource vector and a negative sample resource vector respectively according to the positive sample resource features and the negative sample resource features through the resource neural network of the resource recall model; Determining a first matching degree between the user feature vector and the positive sample resource vector, and determining a second matching degree between the user feature vector and the negative sample resource vector; The preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree.
4. The resource recall model updating method according to claim 3, characterized in that: The preset loss function includes a first difference term, a second difference term and a maximum value term; The preset loss function optimizes the parameters in the resource recall model based on the first matching degree and the second matching degree, including: Determine a first difference between the second matching degree and the first matching degree through the first difference item; Determine a second difference between the first difference and a preset fixed value by using the second difference item; Taking the maximum value between the second difference and the value zero through the maximum value item, and taking the maximum value as the function value of the preset loss function; The parameters in the resource recall model are adjusted in a direction to minimize the function value until a training end condition is met.
5. The resource recall model updating method according to claim 1, characterized in that: The method further comprises: In response to a resource push request from a target user, determining a candidate resource set matching the target user according to the directional information of each resource in the resource library; Determining target user characteristics of the target user and candidate resource characteristics of each candidate resource in the candidate resource set; Inputting the target user characteristics and the candidate resource characteristics of each candidate resource into the updated resource recall model to obtain the matching degree between the target user and each candidate resource; Determining at least one target candidate resource from the candidate resource set according to the matching degree between the target user and each candidate resource; A return result corresponding to the resource push request is determined according to the at least one target candidate resource, where the return result includes a target resource to be displayed.
6. The resource recall model updating method according to claim 5, characterized in that: After determining a return result corresponding to the resource push request according to the at least one target candidate resource, the method further includes: In response to a click operation of the target user on the target resource, acquiring a user identifier of the target user and a resource identifier of the target resource; A resource click log corresponding to the next model update cycle is generated according to the user identifier of the target user and the resource identifier of the target resource.
7. A resource recall model updating device, characterized in that: include: A click log acquisition unit is configured to execute acquisition of resource click logs corresponding to the current model update cycle; A first determining unit is configured to determine a click resource pool according to the resource click log, wherein the click resource pool includes at least one click resource, and the click resource represents a resource clicked by a user in a first historical time period, wherein the first historical time period is a time period between a current model update cycle and a previous model update cycle; a negative sampling unit, configured to select a first click resource in the click resource pool, take the first click resource as a positive sample resource, and determine a negative sample resource corresponding to the positive sample resource from the remaining click resources; the first click resource is any click resource in the click resource pool, and the remaining click resources are click resources in the click resource pool other than the first click resource; A training data generating unit is configured to generate training samples in a training sample set according to positive sample resources, negative sample resources corresponding to the positive sample resources, and users who click on the positive sample resources; A model updating unit is configured to optimize the parameters in the resource recall model using a preset loss function according to the training sample set to obtain an updated resource recall model; the preset loss function is determined based on the degree of matching between the user and the positive sample resources in the training sample and the degree of matching between the user and the negative sample resources.
8. The resource recall model updating device according to claim 7, characterized in that: The training data generating unit comprises: A first determining unit is configured to determine a user who clicks the positive sample resource and obtain a user feature of the user; A second determining unit is configured to determine a positive sample resource feature of the positive sample resource; A third determining unit is configured to determine a negative sample resource feature of a negative sample resource corresponding to the positive sample resource; The generating subunit is configured to execute the formation of training samples in a training sample set by using the user features, the positive sample resource features and the negative sample resource features.
9. The resource recall model updating device according to claim 8, characterized in that: The model updating unit comprises: An input unit is configured to execute, for each training sample in the training sample set, inputting the user features, positive sample resource features, and negative sample resource features included in the training sample into the resource recall model; A first network unit is configured to execute a user neural network through the resource recall model and obtain a user feature vector according to the user feature; A second network unit is configured to execute a resource neural network through the resource recall model, and obtain a positive sample resource vector and a negative sample resource vector according to the positive sample resource feature and the negative sample resource feature respectively; A matching unit, configured to determine a first matching degree between the user feature vector and the positive sample resource vector, and determine a second matching degree between the user feature vector and the negative sample resource vector; The optimization unit is configured to execute a preset loss function to optimize the parameters in the resource recall model based on the first matching degree and the second matching degree.
10. The resource recall model updating device according to claim 9, characterized in that: The preset loss function includes a first difference term, a second difference term and a maximum value term; the optimization unit includes: A fourth determining unit is configured to determine a first difference between the second matching degree and the first matching degree by using the first difference item; A fifth determining unit is configured to determine a second difference between the first difference and a preset fixed value by using the second difference item; A sixth determining unit is configured to perform the step of taking the maximum value between the second difference and the value zero through the maximum value item, and taking the maximum value as the function value of the preset loss function; The parameter adjustment unit is configured to adjust the parameters in the resource recall model in a direction of minimizing the function value until a training end condition is met.
11. The resource recall model updating device according to claim 7, characterized in that: The device also includes: A request response unit, configured to execute a response to a resource push request of a target user, and determine a candidate resource set matching the target user according to the directional information of each resource in the resource library; a seventh determining unit, configured to determine a target user feature of the target user and a candidate resource feature of each candidate resource in the candidate resource set; a matching degree prediction unit, configured to input the target user characteristics and the candidate resource characteristics of each candidate resource into the updated resource recall model to obtain the matching degree between the target user and each candidate resource; An eighth determining unit is configured to determine at least one target candidate resource from the candidate resource set according to a matching degree between the target user and each candidate resource; The ninth determining unit is configured to determine a return result corresponding to the resource push request according to the at least one target candidate resource, wherein the return result includes a target resource to be displayed.
12. The resource recall model updating device according to claim 11, characterized in that: The device also includes: an acquisition unit, configured to acquire a user identifier of the target user and a resource identifier of the target resource in response to a click operation of the target user on the target resource; The click log generating unit is configured to generate a resource click log corresponding to the next model update cycle according to the user identifier of the target user and the resource identifier of the target resource.
13. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the resource recall model updating method according to any one of claims 1 to 6. 14 . A storage medium, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the resource recall model updating method according to any one of claims 1 to 6.
