A media resource recall method, device, server and storage medium

Abstract

The present disclosure relates to a media resource recall method, device, server and storage medium. The method includes: obtaining at least one seed media resource corresponding to user information associated with a client; determining at least one candidate media resource corresponding to each seed media resource, taking the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, and connecting every two vertices to generate an undirected graph; dividing the undirected graph according to the attribute information of each edge in the undirected graph to obtain at least two subgraphs, and determining the recalled media resource from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph. The solution disclosed in the present disclosure solves the problem in the related art that the recall frequency of popular media resources is high, but the recall frequency of unpopular media resources is low. While improving the diversity of recalled media resources, it also improves the recall frequency of unpopular media resources.

Description

Technical Field

[0001] The present disclosure relates to computer technology, and more particularly to a method, device, server, and storage medium for recalling media resources. Background Art

[0002] The development of computer and network technologies has made information more and more developed and its dissemination more and more convenient. Accordingly, how to select more targeted content from the vast amount of information and content to present to users has become a problem worthy of attention.

[0003] Media resources are a relatively special type of information and content. Current internet media resource systems generally consist of two phases: retrieval and sorting. Retrieval is responsible for acquiring a large number of media resources from the system's entire database that match user information. Related technologies primarily use model-based or collaborative filtering methods to retrieve media resources. However, the diversity of media resources retrieved using these two methods is limited, with both methods favoring popular media resources over less popular ones. Summary of the Invention

[0004] The present disclosure provides a media resource recall method, device, server, and storage medium to at least solve the problem in related technologies that popular media resources are recalled frequently but unpopular media resources are recalled frequently.

[0005] The technical solutions of the embodiments of the present disclosure are as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, a method for recalling a media resource is provided, comprising:

[0007] In response to a media resource playback instruction triggered by a client, obtaining at least one seed media resource corresponding to user information associated with the client, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening;

[0008] Determining at least one candidate media resource corresponding to each of the seed media resources, using the media resource vector of each media resource in the media resource set consisting of the seed media resources and the candidate media resources as a vertex, connecting every two vertices to generate an undirected graph, wherein attribute information of an edge between every two vertices in the undirected graph is determined based on the similarity between the corresponding media resource vectors;

[0009] The undirected graph is divided into at least two subgraphs according to the attribute information of each edge in the undirected graph, and the recalled media resources are determined from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains the media resource vector of at least one media resource in the media resource set.

[0010] Optionally, in response to a media resource playback instruction triggered by the client, the step of acquiring at least one seed media resource corresponding to user information associated with the client includes:

[0011] In response to a media resource playback instruction triggered by a client, obtaining user information associated with the client;

[0012] At least one seed media resource corresponding to the user information is selected from the historical media resources recorded in the target media resource queue entering the media resource system, wherein the target media resource queue stores historical media resources whose media resource value parameters exceed a first set threshold.

[0013] Optionally, the step of determining at least one candidate media resource corresponding to each seed media resource includes:

[0014] At least one media resource vector of the candidate media resource similar to each of the seed media resources is retrieved from the media resource vector library.

[0015] Optionally, the step of using the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resources as a vertex and connecting every two vertices to generate an undirected graph includes:

[0016] Generate a vertex set based on the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resources;

[0017] The target vertex in the vertex set is connected to all vertices except the target vertex respectively to generate the undirected graph.

[0018] Optionally, the step of dividing the undirected graph into at least two subgraphs according to the attribute information of each edge in the undirected graph includes:

[0019] Deleting edges in the undirected graph whose attribute information is less than a second set threshold;

[0020] Each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by segmenting the undirected graph.

[0021] Optionally, the step of determining the recalled media resource from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph includes:

[0022] Calculating a weighted average of the media resource vectors included in each subgraph, and using the weighted average as the center point vector of each subgraph;

[0023] Calculating the distance between each media resource vector in each of the subgraphs and the center point vector of the corresponding subgraph;

[0024] When the distance is less than a third set threshold, the seed media resource or the candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

[0025] Optionally, the media resource vector is determined by a pre-trained delivery model;

[0026] The delivery model is trained based on positive samples consisting of media resources in the target media resource queue that have participated in value screening and have been delivered, and negative samples consisting of media resources in the target media resource queue that have participated in value screening but have not been delivered;

[0027] The following steps are used to determine the media resources in the target media resource queue:

[0028] A target recalled media resource whose media resource value parameter among the recalled media resources corresponding to the user information exceeds a set threshold is determined, and the target recalled media resource is added to a target media resource queue.

[0029] According to a second aspect of an embodiment of the present disclosure, a media resource recall device is provided, including:

[0030] an acquisition module configured to, in response to a media resource playback instruction triggered by a client, acquire at least one seed media resource corresponding to user information associated with the client, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening;

[0031] a determination module configured to determine at least one candidate media resource corresponding to each seed media resource, use the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, connect every two vertices to generate an undirected graph, and determine attribute information of an edge between every two vertices in the undirected graph based on a similarity between the corresponding media resource vectors;

[0032] A segmentation module is configured to segment the undirected graph into at least two subgraphs based on the attribute information of each edge in the undirected graph, and determine the recalled media resources from the media resource set based on the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains the media resource vector of at least one media resource in the media resource set.

[0033] Optionally, the acquisition module is specifically configured to:

[0034] In response to a media resource playback instruction triggered by a client, obtaining user information associated with the client;

[0035] At least one seed media resource corresponding to the user information is selected from the historical media resources recorded in the target media resource queue entering the media resource system, wherein the target media resource queue stores historical media resources whose media resource value parameters exceed a first set threshold.

[0036] Optionally, the determination module includes: a retrieval submodule;

[0037] The retrieval submodule is configured to retrieve, from a media resource vector library, a media resource vector of at least one candidate media resource that is similar to each seed media resource.

[0038] Optionally, the determination module further includes: a generation submodule;

[0039] The generating submodule is configured to generate a vertex set based on the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resources;

[0040] The target vertex in the vertex set is connected to all vertices except the target vertex respectively to generate the undirected graph.

[0041] Optionally, the segmentation module includes: a segmentation submodule;

[0042] The segmentation submodule is configured to delete the edges in the undirected graph whose attribute information is less than a second set threshold;

[0043] Each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by segmenting the undirected graph.

[0044] Optionally, the segmentation module further includes: a determination submodule;

[0045] The determining submodule is configured to calculate a weighted average value of the media resource vectors included in each subgraph, and use the weighted average value as the center point vector of each subgraph;

[0046] Calculating the distance between each media resource vector in each of the subgraphs and the center point vector of the corresponding subgraph;

[0047] When the distance is less than a third set threshold, the seed media resource or the candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

[0048] Optionally, the media resource vector is determined by a pre-trained delivery model;

[0049] The delivery model is trained based on positive samples consisting of media resources in the target media resource queue that have participated in value screening and have been delivered, and negative samples consisting of media resources in the target media resource queue that have participated in value screening but have not been delivered;

[0050] The media resources in the target media resource queue are determined by a second determining module:

[0051] The second determining module is configured to determine a target recalled media resource whose media resource value parameter exceeds a set threshold among the recalled media resources corresponding to the user information, and add the target recalled media resource to a target media resource queue.

[0052] According to a third aspect of an embodiment of the present disclosure, a server is provided, including:

[0053] processor;

[0054] a memory for storing executable instructions for the processor;

[0055] The processor is configured to execute the instructions to implement the media resource recall method as described in any embodiment of the present disclosure.

[0056] 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 a server, the server is enabled to execute the media resource recall method as described in any embodiment of the present disclosure.

[0057] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided. When instructions in the computer program product are executed by a processor of a server, the media resource recall method described in any embodiment of the present disclosure is implemented.

[0058] The technical solutions provided by the embodiments of the present disclosure bring at least the following beneficial effects: by obtaining at least one seed media resource corresponding to user information associated with a client, it is possible to ensure that multiple seed media resources corresponding to the user information are obtained; by determining at least one candidate media resource corresponding to each seed media resource, even if the candidate media resource is an unpopular media resource, it can be obtained; the media resource vector of each media resource in the media resource set consisting of the seed media resources and the candidate media resources is used as a vertex, and every two vertices are connected to generate an undirected graph; the undirected graph is partitioned according to the attribute information of each edge in the undirected graph to obtain at least two subgraphs, and the recalled media resources are determined from the media resource set based on the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph, which can ensure the diversity of the recalled media resources and can include both popular media resources and unpopular media resources; therefore, the solution of the embodiments of the present disclosure can solve the problem in the related art that the recall frequency of popular media resources is high, but the recall frequency of unpopular media resources is low. While improving the diversity of the recalled media resources, it also improves the recall frequency of unpopular media resources.

[0059] 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 disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying 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 an improper limitation of the present disclosure.

[0061] Figure 1 The figure is a flowchart of a method for recalling media resources according to an exemplary embodiment.

[0062] Figure 2 The figure is a flowchart of a method for recalling media resources according to an exemplary embodiment.

[0063] Figure 3 The figure is a flowchart of a method for recalling media resources according to an exemplary embodiment.

[0064] Figure 4 is an undirected graph including four vertices according to an exemplary embodiment.

[0065] Figure 5 The figure is a flowchart of a method for recalling media resources according to an exemplary embodiment.

[0066] Figure 6 The figure is a flowchart of a method for recalling media resources according to an exemplary embodiment.

[0067] Figure 7The figure is a workflow diagram of a media resource system according to an exemplary embodiment.

[0068] Figure 8 The figure is a block diagram of a media resource recall device according to an exemplary embodiment.

[0069] Figure 9 The figure is a block diagram of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0070] 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 with reference to the accompanying drawings.

[0071] It should be noted that the terms "first," "second," and the like 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 precedence. It should be understood that the numbers used in this manner are interchangeable 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 apparatus and methods consistent with certain aspects of the present disclosure as detailed in the appended claims.

[0072] Figure 1 FIG. 1 is a flow chart showing a method for recalling media resources according to an exemplary embodiment. Figure 1 As shown, the media resource recall method can be executed by a media resource recall device, which can be implemented by software and / or hardware and used in a server. The media resource recall method includes the following steps.

[0073] In step S11 , in response to a media resource play instruction triggered by the client, at least one seed media resource corresponding to user information associated with the client is obtained.

[0074] It should be noted that the media resources involved in this embodiment may be print resources, Internet resources, or advertising resources, etc., which are not limited in this embodiment.

[0075] The client, also known as the user end, refers to a program that provides local services to a user; for example, a video playback program, a chat program, or a weather query program, etc., which are not limited in this embodiment. In an optional implementation of this embodiment, when a user opens the client or browses information within the client, a playback instruction of a media resource may be triggered.

[0076] The user information associated with the client may include the user ID (Identity Document, identity identification number) of the client, the user name or the user's mobile phone number, etc., which are not limited in this embodiment.

[0077] In an optional implementation of this embodiment, after receiving a media resource playback instruction triggered by a client, the server may further obtain at least one seed media resource corresponding to the user information associated with the client; wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening (e.g., bidding). For example, at least one seed media resource may be selected from historical media resources associated with the user's ID that participated in the last bidding; or at least one seed media resource may be selected from historical media resources associated with the user's mobile phone number that participated in the last bidding, which is not limited in this embodiment.

[0078] For example, after receiving the media resource playback instruction triggered by the client, the server can obtain the 100 historical media resources that participated in the last bidding and are associated with the user name of the user who logged into the client in the media resource system, and determine these 100 historical media resources as seed media resources corresponding to the user information associated with the client.

[0079] In step S12, at least one candidate media resource corresponding to each seed media resource is determined, the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource is used as a vertex, and every two vertices are connected to generate an undirected graph.

[0080] Among them, the media resource vector is the result of operation obtained by using a pre-trained delivery model to operate on the seed media resource or the candidate media resource. For example, the seed media resource or the candidate media resource is input into the pre-trained delivery model for prediction, thereby obtaining the media resource vector corresponding to the seed media resource or the candidate media resource. In an optional implementation of this embodiment, the delivery model can be trained based on positive samples consisting of media resources in the target media resource queue that have participated in value screening and have been delivered, and negative samples consisting of media resources in the target media resource queue that have participated in value screening but have not been delivered; wherein, the following steps can be used to determine the media resources in the target media resource queue: determine the target recall media resource whose media resource value parameter in the recall media resource corresponding to the user information exceeds the set threshold, and add the target recall media resource to the target media resource queue.

[0081] The candidate media resource is similar to its corresponding seed media resource; optionally, if the media resource vectors of two media resources are similar, then the two media resources are determined to be similar. For example, when the distance between the media resource vectors of two media resources is less than a set similarity threshold, the two media resources are determined to be similar. For example, if the seed media resource is the media resource of daily life product A produced by manufacturer M, then the corresponding candidate media resource can be the media resource of daily life product B produced by manufacturer M; it can also be the media resource of daily life product A produced by manufacturer N, which is not limited in this embodiment.

[0082] It should also be noted that the candidate media resources involved in this embodiment may include both popular media resources and unpopular media resources corresponding to the seed media resources. It can be understood that as long as the media resources are similar to the seed media resources, they can be obtained. This embodiment does not limit parameters such as the category or value of the candidate media resources.

[0083] In an optional implementation of this embodiment, media resource vectors of multiple candidate media resources that are similar to the media resource vector of each seed media resource can be retrieved from the media resource vector library. The media resource vector library includes the association relationship between media resources and corresponding media resource vectors. Based on the association relationship between the media resource vectors and media resources in the media resource vector library, candidate media resources are retrieved from the media resource library to obtain a media resource set including seed media resources and candidate media resources. For example, based on the media resource vector of seed media resource A, 100 media resource vectors of candidate media resources similar to seed media resource A can be obtained from the media resource library; or 200 media resource vectors of candidate media resources similar to seed media resource B can be obtained from the media resource library, which is not limited in this embodiment. The media resource vector of the seed media resource is the result of calculation obtained by calculating the seed media resource using a pre-trained distribution model.

[0084] In an optional implementation of this embodiment, after determining at least one candidate media resource corresponding to each seed media resource, these seed media resources and candidate media resources can be added to a media resource set. Furthermore, each media resource vector in the media resource set can be used as a vertex, and then every two vertices can be connected to generate an undirected graph.

[0085] The attribute information of the edge between each two vertices in the undirected graph is determined based on the similarity between the media resource vectors. For example, if vertex A corresponds to media resource vector A and vertex B corresponds to media resource vector B, then the attribute information of the edge between vertex A and vertex B is the similarity between media resource vector A and media resource vector B.

[0086] For example, if 10 seed media resources corresponding to the user information associated with the client are obtained, and 20 candidate media resources similar to each seed media resource are obtained in the media resource library, then these 10 seed media resources and 200 candidate media resources can be added to the media resource collection; further, all media resource vectors in the media resource collection can be used as vertices of an undirected graph, and every two vertices can be connected to generate an undirected graph, and the attribute information of the edge between every two vertices is the similarity between the corresponding media resource vectors.

[0087] In step S13, the undirected graph is divided into at least two subgraphs according to the attribute information of each edge in the undirected graph, and the recalled media resources are determined from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph.

[0088] In which, each subgraph includes a media resource vector of at least one media resource in the media resource set; for example, it may include a media resource vector of at least one seed media resource in the media resource set, a media resource vector of a candidate media resource, or simultaneously include a media resource vector of at least one seed media resource in the media resource set and a media resource vector of at least one candidate media resource, which is not limited in this embodiment.

[0089] In an optional implementation of this embodiment, after taking the media resource vector of each media resource in the media resource set as a vertex and connecting every two vertices to generate an undirected graph, the generated undirected graph can be segmented according to the attribute information of each edge in the undirected graph. For example, the edges in the undirected graph whose similarity (attribute information of the edge) is the same or the difference is less than a set threshold (for example, 0.01) can be deleted and segmented into a subgraph.

[0090] Furthermore, the center point vector of each subgraph can be calculated. For example, in the subgraph A obtained by segmentation, all media resource vectors contained in subgraph A can be added and then averaged, and the average value can be used as the center point vector of subgraph A. After the center point vector of each subgraph is calculated, the recalled media resources can be determined from the media resource set based on the distance between the media resource vector in each subgraph and the center point vector of the corresponding subgraph. For example, in the above example, after the center point vector of subgraph A is calculated, the distance between each media resource vector in subgraph A and the center point vector can be calculated. When the distance value is less than the third set threshold, the seed media resource or candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

[0091] The third set threshold may be a value such as 0.1, 0.2 or 0.05, which is not limited in this embodiment.

[0092] In the related art, media resources are mainly recalled based on model-based methods or collaborative filtering-based methods. Among them, the model-based method: the media resources sent to the client are used as negative samples, and the media resources clicked by the user on the client are used as positive samples, the model is trained, and the recalled media resources are determined based on the trained model. The collaborative filtering-based method: mining is performed based on the association rules between users and users, media resources and media resources, so as to obtain the recalled media resources. The methods in the related art tend to recall popular media resources with higher media resource value parameters, and it is difficult to recall unpopular media resources with lower media resource value parameters, resulting in poor diversity of recalled media resources.

[0093] The solution of this embodiment ensures that multiple seed media resources corresponding to user information are obtained by obtaining at least one seed media resource corresponding to user information associated with the client; by determining at least one candidate media resource corresponding to each seed media resource, even if the candidate media resource is an unpopular media resource (a media resource with a low media resource value parameter), it can be obtained; the media resource vector of each media resource in the media resource set consisting of seed media resources and candidate media resources is used as a vertex, and every two vertices are connected to generate an undirected graph; the undirected graph is divided according to the attribute information of each edge in the undirected graph to obtain at least two subgraphs, and the recalled media resources are determined from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph, which can ensure the diversity of the recalled media resources (including both popular media resources and unpopular media resources); therefore, the solution of the embodiment of the present disclosure can solve the problem in the related art that the recall frequency of popular media resources is high, but the recall frequency of unpopular media resources is low. While improving the diversity of recalled media resources, it also improves the recall frequency of unpopular media resources.

[0094] Figure 2 This is a flow chart of a media resource recall method according to an exemplary embodiment. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 2 As shown, the media resource recall method includes the following steps.

[0095] In step S21, in response to a media resource play instruction triggered by the client, user information associated with the client is obtained; and at least one seed media resource corresponding to the user information is selected from the recorded historical media resources in the target media resource queue entering the media resource system.

[0096] The target media resource queue stores historical media resources whose media resource value parameters exceed a first set threshold. The media resource value parameter is CPM (Cost Per Mille), which is the cost per thousand impressions of each media resource. The first set threshold can be 10 yuan, 100 yuan, or 200 yuan, etc., and is not limited in this embodiment.

[0097] In an optional implementation of this embodiment, after receiving the media resource playback instruction triggered by the client, the server can further obtain user information associated with the client; for example, it can obtain the user ID, user name or account number of the logged-in client, which is not limited in this embodiment.

[0098] Furthermore, at least one seed media resource corresponding to the user information can be selected from the historical media resources recorded in the target media resource queue of the media resource system. For example, if the time when the server receives the media resource playback instruction triggered by the client is July 2, 2020, the historical media resources recorded in the target media resource queue can be 100 media resources participating in the bidding associated with the user information on July 1, 2020. It should be noted that the number of historical media resources selected can be an empirical value related to the application scenario, and the more historical media resources selected, the greater the selection space for seed media resources.

[0099] In step S22, a media resource vector of at least one candidate media resource similar to each seed media resource is retrieved from the media resource vector library.

[0100] In an optional implementation of this embodiment, after selecting at least one seed media resource corresponding to the user information from the historical media resources recorded in the target media resource queue entering the media resource system, the media resource vector library can be further retrieved to obtain the media resource vector of at least one candidate media resource similar to each seed media resource.

[0101] The media resource vector library stores media resource vectors for all media resources. This term "all media resources" refers to media resources for all products, individuals, or companies, and is not limited to this term in this embodiment. It should be noted that the media resource vectors for all media resources can be calculated using a pre-trained delivery model for each media resource. For example, each media resource can be sequentially input into the pre-trained delivery model for prediction, thereby obtaining a media resource vector for each media resource and storing these media resource vectors in the media resource vector library.

[0102] In an optional implementation of this embodiment, a media resource vector of at least one candidate media resource similar to each seed media resource can be retrieved from the media resource vector library using a vector retrieval algorithm such as HNSW (Hierarchical Navigable Small World) or LSH (Locality Sensitive Hashing).

[0103] In step S23 , the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resources is used as a vertex, and every two vertices are connected to generate an undirected graph.

[0104] In step S24, the undirected graph is divided into at least two subgraphs according to the attribute information of each edge in the undirected graph, and the recalled media resources are determined from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph.

[0105] The solution of this embodiment, after selecting at least one seed media resource corresponding to user information from the historical media resources recorded in the target media resource queue entering the media resource system, can further retrieve the media resource vector of at least one candidate media resource similar to each seed media resource from the media resource vector library. The retrieved media resource vectors include both popular media resources and unpopular media resources. This ensures that diverse candidate media resources are obtained, even unpopular media resources can be obtained, providing a basis for subsequently improving the diversity of recalled media resources.

[0106] Figure 3 This is a flow chart of a media resource recall method according to an exemplary embodiment. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 3 As shown, the media resource recall method includes the following steps.

[0107] In step S31, in response to a media resource play instruction triggered by the client, user information associated with the client is obtained; and at least one seed media resource corresponding to the user information is selected from the recorded historical media resources in the target media resource queue entering the media resource system.

[0108] In step S32, a media resource vector of at least one candidate media resource similar to each seed media resource is retrieved from the media resource vector library.

[0109] In step S33, the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource is used as a vertex, and every two vertices are connected to generate an undirected graph. In an optional implementation of this embodiment, the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource is used as a vertex, and every two vertices are connected to generate an undirected graph, which may include: generating a vertex set based on the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource; and connecting the target vertex in the vertex set with all vertices except the target vertex to generate an undirected graph. The target vertex can be any media resource vector in the media resource set, which is not limited in this embodiment.

[0110] In an optional implementation, after obtaining the candidate media resources corresponding to all seed media resources, the media resource vector of each media resource in the media resource set can be further used as a vertex set, that is, the media resource vector of each media resource in the media resource set can be used as each vertex of an undirected graph, wherein the media resource set includes all seed media resources and candidate media resources corresponding to all seed media resources; the similarity of the media resource vectors between every two media resources in the media resource set can be used as the attribute information of each edge, that is, the similarity of the media resource vectors between every two media resources in the media resource set can be used as the attribute information of each edge of the undirected graph.

[0111] For example, if the media resource set includes 4 media resources (seed media resources and candidate media resources), the generated undirected graph can be as follows: Figure 4 As shown, it includes 4 vertices (1, 2, 3 and 4) and 6 edges (5, 6, 7, 8, 9 and 10), where the vertices can represent the media resource vectors of the 4 media resources; the attribute information of the edges can represent the similarity of the media resource vectors between each two media resources in the 4 media resources.

[0112] In step S34, edges in the undirected graph whose attribute information is less than a second set threshold are deleted; each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by splitting the undirected graph.

[0113] In an optional implementation of this embodiment, after taking the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex and connecting every two vertices to generate an undirected graph, the edges in the undirected graph whose attribute information (similarity) is less than a second set threshold can be further deleted; wherein, the second set threshold can be a value such as 0.2, 0.3 or 0.5, which is not limited in this embodiment.

[0114] For example, if the similarity between the first media resource vector and the second media resource vector is 0.1 (the second threshold is 0.2), the edge corresponding to the media resource vector similarity (attribute information) between the first media resource vector and the second media resource vector can be deleted from the undirected graph. In this case, the first media resource corresponding to the first media resource vector and the second media resource corresponding to the second media resource vector are no longer connected. This setting provides a basis for subsequent segmentation of the undirected graph and reduces computational complexity.

[0115] It is understandable that after deleting the edges in the undirected graph whose similarity is less than the second set threshold, the undirected graph can be converted from a connected graph to multiple connected undirected graphs. At this time, each vertex of the undirected graph can be traversed to determine the multiple connected subgraphs included in the undirected graph. It should be noted that if multiple connected undirected graphs (for example, 2) are not obtained after deleting the edges in the undirected graph whose similarity is less than the second set threshold, the second set threshold can be adjusted at this time. For example, the second set threshold can be reduced, for example, the second set threshold can be adjusted from 0.2 to 0.15, etc., until at least two connected undirected graphs are obtained.

[0116] For example, Figure 4 The undirected graph shown is processed, and the edges in the undirected graph whose similarity is less than the second set threshold are deleted. Assuming that edges 5, 6 and 7 are deleted, then by further traversing each vertex of the undirected graph, two subgraphs can be determined. The first subgraph includes vertex 1, and the second subgraph includes vertex 2, vertex 3 and vertex 4.

[0117] In step S35 , the recalled media resources are determined from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph.

[0118] The solution of this embodiment generates a vertex set based on the media resource vector of each media resource in the media resource set composed of seed media resources and candidate media resources; connects the target vertex in the vertex set with all vertices except the target vertex to generate an undirected graph; deletes the edges in the undirected graph whose attribute information is less than a second set threshold; traverses each vertex of the undirected graph to determine at least two connected subgraphs in the undirected graph, and the at least two connected subgraphs are at least two subgraphs obtained by splitting the undirected graph. The undirected graph can be quickly split, thereby improving the efficiency of media resource recall, and can quickly respond to media resource playback instructions triggered by the client, reducing the time it takes for the media resource system to send media resources.

[0119] Figure 5This is a flow chart of a media resource recall method according to an exemplary embodiment. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 5 As shown, the media resource recall method includes the following steps.

[0120] In step S51, in response to a media resource play instruction triggered by the client, at least one seed media resource corresponding to user information associated with the client is obtained.

[0121] In step S52 , at least one candidate media resource corresponding to each seed media resource is determined, and the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource is used as a vertex. Every two vertices are connected to generate an undirected graph.

[0122] In step S53, the edges in the undirected graph whose attribute information is less than the second set threshold are deleted; each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by splitting the undirected graph.

[0123] In step S54, the weighted average of the media resource vectors included in each subgraph is calculated, and the weighted average is used as the center point vector of each subgraph; the distance between the media resource vectors in each subgraph and the center point vector of the corresponding subgraph is calculated; when the distance is less than a third set threshold, the seed media resource or candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined to be the media resource to be recalled.

[0124] In an optional implementation of this embodiment, after the generated undirected graph is divided into at least two subgraphs, the weighted average of the media resource vectors included in each subgraph can be further calculated, and the calculated weighted average is used as the center point vector of each subgraph.

[0125] For example, if the segmented subgraph A includes 10 media resource vectors (including seed media resource vectors and candidate media resource vectors), different weights can be set for each of the 10 media resource vectors. For example, the weight of the media resource vector at the center of the subgraph is set to 1, and the weight of the media resource vector at the boundary of the subgraph is set to 0.8, etc. This is not restricted in this embodiment; further, the weighted average of the 10 media resource vectors can be calculated, and the weighted average of the 10 media resource vectors can be determined as the center point vector of subgraph A.

[0126] Furthermore, the distance between all media resource vectors in subgraph A and the calculated center point vector of subgraph A can be calculated. For example, the distance between 10 media resource vectors in subgraph A and the center point vector can be calculated to obtain 10 distance values. Furthermore, it is determined whether the distance is less than a third predetermined threshold. The third predetermined threshold can be a value such as 0.1, 0.2, or 0.05, and is not limited in this embodiment. For example, if all 10 distance values ​​are less than the third predetermined threshold, then the 10 media resource vectors in subgraph A can be determined as media resources to be recalled.

[0127] In another optional implementation of this embodiment, the median or average value of each media resource vector included in each subgraph can also be calculated, and the calculated median or average value can be used as the center point vector of each subgraph; the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph is calculated; when the distance is less than a third set threshold, the seed media resource or candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined to be the media resource to be recalled. The solution of this embodiment, by calculating the weighted average value of each media resource vector included in each subgraph, and using the weighted average value as the center point vector of each subgraph; calculating the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; when the distance is less than a third set threshold, the seed media resource or candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined to be the media resource to be recalled, can improve the diversity of the media resources to be recalled (both hot media resources and hot media resources can be obtained).

[0128] In order to enable those skilled in the art to better understand the media resource recall method involved in this embodiment, a specific example is used below for illustration. It should be noted that in this embodiment, advertisements are used as media resources for illustration; Figure 6 As shown, the specific process includes:

[0129] In step S61, the user seed advertisement is obtained.

[0130] Each advertisement that enters the bidding is recorded in the advertising system, and each user randomly retains M advertisements that enter the bidding queue, where M can be any value and is not limited in this embodiment.

[0131] In step S62, similar advertisements are retrieved.

[0132] The seed advertisement vectors are obtained by issuing the model, and a vector search algorithm such as HNSW or LSH is used to retrieve at least one advertisement similar to each seed advertisement.

[0133] In step S63, an undirected graph is generated.

[0134] In an advertisement set consisting of various sub-advertisements and at least one advertisement similar to the various sub-advertisements, the similarity between every two advertisement vectors is calculated, an undirected graph is generated, and edges with similarities less than a set threshold are deleted.

[0135] In step S64, the undirected graph is minimally cut to obtain at least two subgraphs.

[0136] A minimum cut algorithm is used to partition an undirected graph into at least two subgraphs. Each subgraph serves as a focus point, and the center point vector is the weighted average of the vectors of each dimension of each point in the subgraph. It should be noted that the minimum cut algorithm involved in this embodiment can also be replaced by a clustering algorithm, which is not limited in this embodiment.

[0137] In step S65, advertisements are screened.

[0138] Select at least one advertisement vector from each subgraph that is closest to the center point.

[0139] In step S66, the recall result is returned.

[0140] The solution of this embodiment uses the center point of the advertisement subgraph as the point of interest, which can solve the diversity problem to a certain extent. It adopts the minimum cut algorithm and vector retrieval algorithm to improve computing efficiency. The seed point selects the advertisements entering the bidding queue as the seed point, ensuring sufficient diversity and solving the problems of slow feedback time and insufficient diversity in related technologies that use user behavior as seed points.

[0141] Figure 7 This is a workflow diagram of an advertising system according to an exemplary embodiment. Figure 7 ,The workflow of the advertising system mainly includes the following steps.

[0142] In step S71, recall.

[0143] According to the advertisement recall method involved in each of the above embodiments, advertisements that meet user requests are recalled.

[0144] In step S72, sorting is performed.

[0145] Based on user information, ad bids, and ad information, a CPM estimate is made for each recalled ad.

[0146] In step S73, bidding is performed.

[0147] Based on the CPM estimation results, filter out low-value ads and select at least one high-value ad.

[0148] In step S74, it is issued.

[0149] At least one high-value advertisement obtained through screening is sent to the client, and the sent advertisement is displayed on the client.

[0150] In this embodiment, during the bidding and delivery phases, the delivery model is trained using positive samples (advertisements that have been bid on and delivered) and negative samples (advertisements that have been bid on but not delivered). The delivery model used in this embodiment utilizes supervised training, taking into account the inherent value of the ads. Furthermore, as a server-side model, it avoids the slow feedback issues of client-side models and provides rapid feedback on client-triggered ad playback requests.

[0151] Figure 8 FIG. 1 is a block diagram of a media resource recall device according to an exemplary embodiment. Figure 8 The device includes a detection module 81, a determination module 82 and a segmentation module 83.

[0152] An acquisition module 81 is configured to, in response to a media resource playback instruction triggered by a client, acquire at least one seed media resource corresponding to user information associated with the client, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening;

[0153] Determination module 82 is configured to determine at least one candidate media resource corresponding to each seed media resource, use the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, connect every two vertices to generate an undirected graph, and determine the attribute information of the edge between every two vertices in the undirected graph based on the similarity between the corresponding media resource vectors;

[0154] The segmentation module 83 is configured to segment the undirected graph into at least two subgraphs based on the attribute information of each edge in the undirected graph, and determine the recalled media resources from the media resource set based on the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains at least one media resource in the media resource set.

[0155] The solution of this embodiment is to obtain at least one seed media resource corresponding to user information associated with the client in response to a media resource playback instruction triggered by the client through an acquisition module; determine at least one candidate media resource corresponding to each seed media resource through a determination module, use the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, and connect every two vertices to generate an undirected graph; and use the segmentation module to segment the undirected graph according to the attribute information of each edge in the undirected graph to obtain at least two subgraphs, and determine the recalled media resources from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph, thereby solving the problem in related technologies that the recall frequency of popular media resources is high but the recall frequency of unpopular media resources is low. While improving the diversity of recalled media resources, the recall frequency of unpopular media resources is also improved.

[0156] Optionally, the acquisition module 81 is specifically configured to:

[0157] Responding to a media resource playback instruction triggered by the client, obtaining user information associated with the client;

[0158] At least one seed media resource corresponding to the user information is selected from the recorded historical media resources in the target media resource queue entering the media resource system, wherein the target media resource queue stores historical media resources whose media resource value parameters exceed a first set threshold.

[0159] Optionally, the determination module 82 includes: a retrieval submodule;

[0160] The retrieval submodule is configured to retrieve, from the media resource vector library, a media resource vector of at least one candidate media resource that is similar to each seed media resource.

[0161] Optionally, the determination module 82 further includes: a generation submodule;

[0162] a generating submodule configured to generate a vertex set based on a media resource vector of each media resource in a media resource set consisting of the seed media resource and the candidate media resources;

[0163] Connect the target vertex in the vertex set with all vertices except the target vertex to generate an undirected graph.

[0164] Optionally, the segmentation module 83 includes: a segmentation submodule;

[0165] A segmentation submodule configured to delete edges in the undirected graph whose attribute information is less than a second set threshold;

[0166] Each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by partitioning the undirected graph.

[0167] Optionally, the segmentation module 83 further includes: a determination submodule;

[0168] a determination submodule configured to calculate a weighted average of the media resource vectors included in each subgraph, and use the weighted average as the center point vector of each subgraph;

[0169] Calculate the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph;

[0170] When the distance is less than a third set threshold, the seed media resource or the candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

[0171] Optionally, the delivery model is trained based on positive samples consisting of media resources in the target media resource queue that have participated in bidding and have been delivered, and negative samples consisting of media resources in the target media resource queue that have participated in bidding but have not been delivered;

[0172] Among them, the media resource vector is determined by a pre-trained delivery model;

[0173] The delivery model is trained based on positive samples of media resources in the target media resource queue that have participated in value screening and have been delivered, and negative samples of media resources in the target media resource queue that have participated in value screening but have not been delivered.

[0174] The media resources in the target media resource queue are determined by the second determination module:

[0175] The second determining module is configured to determine a target recalled media resource whose media resource value parameter exceeds a set threshold among the recalled media resources corresponding to the user information, and add the target recalled media resource to a target media resource queue.

[0176] Regarding the apparatus 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.

[0177] Figure 9 FIG. 1 is a structural block diagram of a server according to an exemplary embodiment. Figure 9As shown, the server includes a processor 91; a memory 92 for storing executable instructions of the processor 91, and the memory 92 may include RAM (Random Access Memory) and ROM (Read-Only Memory); wherein the processor 91 is configured to execute instructions to implement the above-mentioned media resource recall method;

[0178] That is, in response to a media resource playback instruction triggered by the client, at least one seed media resource corresponding to user information associated with the client is obtained, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening;

[0179] Determine at least one candidate media resource corresponding to each seed media resource, use the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, connect every two vertices to generate an undirected graph, and determine the attribute information of the edge between every two vertices in the undirected graph based on the similarity between the corresponding media resource vectors;

[0180] An undirected graph is divided into at least two subgraphs according to attribute information of each edge in the undirected graph, and a recalled media resource is determined from a media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains at least one media resource in the media resource set.

[0181] In an exemplary embodiment, a storage medium including instructions is also provided, such as a memory 92 storing executable instructions. The above instructions can be executed by a processor 91 of a server (server or smart terminal) to complete the above media resource recall method.

[0182] Alternatively, the storage medium may be a non-transitory computer-readable storage medium, for example, a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0183] In an exemplary embodiment, a computer program product is further provided. When instructions in the computer program product are executed by a processor of a server, the above-mentioned media resource recall method is implemented.

[0184] Other embodiments of the present disclosure will readily occur to those skilled in the art 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 not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0185] 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 can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A media resource recall method, characterized in that: include: In response to a media resource playback instruction triggered by a client, obtaining at least one seed media resource corresponding to user information associated with the client, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening; Determining at least one candidate media resource corresponding to each of the seed media resources, using the media resource vector of each media resource in the media resource set consisting of the seed media resources and the candidate media resources as a vertex, connecting every two vertices to generate an undirected graph, wherein attribute information of an edge between every two vertices in the undirected graph is determined based on the similarity between the corresponding media resource vectors; Splitting the undirected graph according to attribute information of each edge in the undirected graph to obtain at least two subgraphs, and determining the recalled media resource from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains the media resource vector of at least one media resource in the media resource set; The step of determining the recalled media resources from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph comprises: Calculating a weighted average of the media resource vectors included in each subgraph, and using the weighted average as the center point vector of each subgraph; Calculating the distance between each media resource vector in each of the subgraphs and the center point vector of the corresponding subgraph; When the distance is less than a third set threshold, the seed media resource or the candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

2. The method according to claim 1, characterized in that The step of obtaining at least one seed media resource corresponding to user information associated with the client in response to a media resource playback instruction triggered by the client includes: In response to a media resource playback instruction triggered by a client, obtaining user information associated with the client; At least one seed media resource corresponding to the user information is selected from the historical media resources recorded in the target media resource queue entering the media resource system, wherein the target media resource queue stores historical media resources whose media resource value parameters exceed a first set threshold.

3. The method according to claim 1, characterized in that The step of determining at least one candidate media resource corresponding to each seed media resource includes: At least one media resource vector of the candidate media resource similar to each of the seed media resources is retrieved from the media resource vector library.

4. The method according to claim 3, characterized in that The step of using the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex and connecting every two vertices to generate an undirected graph includes: Generate a vertex set based on the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resources; The target vertex in the vertex set is connected to all vertices except the target vertex respectively to generate the undirected graph.

5. The method according to claim 4, characterized in that The step of dividing the undirected graph into at least two subgraphs according to the attribute information of each edge in the undirected graph comprises: Deleting edges in the undirected graph whose attribute information is less than a second set threshold; Each vertex of the undirected graph is traversed to determine at least two connected subgraphs in the undirected graph, where the at least two connected subgraphs are at least two subgraphs obtained by segmenting the undirected graph.

6. The method according to any one of claims 1 to 5, characterized in that The media resource vector is determined by a pre-trained delivery model; The delivery model is trained based on positive samples consisting of media resources in the target media resource queue that have participated in value screening and have been delivered, and negative samples consisting of media resources in the target media resource queue that have participated in value screening but have not been delivered; The following steps are used to determine the media resources in the target media resource queue: A target recalled media resource whose media resource value parameter among the recalled media resources corresponding to the user information exceeds a set threshold is determined, and the target recalled media resource is added to a target media resource queue.

7. A media resource recall device, characterized in that: include: an acquisition module configured to, in response to a media resource playback instruction triggered by a client, acquire at least one seed media resource corresponding to user information associated with the client, wherein the seed media resource is determined based on historical media resources associated with the user information that have participated in value screening; a determination module configured to determine at least one candidate media resource corresponding to each seed media resource, use the media resource vector of each media resource in the media resource set consisting of the seed media resource and the candidate media resource as a vertex, connect every two vertices to generate an undirected graph, and determine attribute information of an edge between every two vertices in the undirected graph based on a similarity between the corresponding media resource vectors; a segmentation module configured to segment the undirected graph according to attribute information of each edge in the undirected graph to obtain at least two subgraphs, and determine the recalled media resources from the media resource set according to the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; wherein each subgraph contains the media resource vector of at least one media resource in the media resource set; The segmentation module further includes: determining submodules; a determination submodule configured to calculate a weighted average of the media resource vectors included in each subgraph, and use the weighted average as the center point vector of each subgraph; Calculate the distance between each media resource vector in each subgraph and the center point vector of the corresponding subgraph; When the distance is less than a third set threshold, the seed media resource or the candidate media resource in the media resource set corresponding to the distance less than the third set threshold is determined as the media resource to be recalled.

8. A server, characterized in that: include: processor; a memory for storing executable instructions for the processor; The processor is configured to execute the instructions to implement the media resource recall method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of a server, the server is enabled to execute the media resource recall method according to any one of claims 1 to 6.

Images