Method, device and storage medium for determining to-be-recommended resource

Abstract

The present disclosure provides a method and device for determining a resource to be recommended, an equipment and a storage medium, relating to the technical field of data processing, in particular to the technical field of big data, information flow, intelligent recommendation and the like, and can be used in the personalized recommendation scene. The specific implementation scheme is as follows: in a plurality of user vectors and a plurality of resource vectors, a user vector of a first user and a resource vector of a first resource are obtained, the plurality of user vectors and the plurality of resource vectors are determined according to resource interaction behavior data of a plurality of users and / or random walk corpus of a user-resource relationship graph; according to the similarity between the user vector of the first user and the resource vector of the first resource, it is determined whether to recall the first resource to obtain a first recalled resource; and according to the recalled resource, a resource to be recommended for the first user is determined, and the recalled resource includes the first recalled resource. Thus, based on the user vector and the resource vector, resource recall is realized, more interesting resources for users are explored, and the information cocoon problem is solved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and more particularly, to a method and device for determining a resource to be recommended, and a storage medium. BACKGROUND

[0002] In resource recommendation technology, there are personalized recommendation technology and non-personalized recommendation technology. The personalized recommendation technology is the core technology and research focus of resource recommendation.

[0003] In related technology, the personalized recommendation technology is mainly based on user interaction behavior data facing resources, such as user click behavior on articles or user playback behavior on videos, to train a personalized recommendation model, and then use the trained personalized recommendation model to predict resources such as music, articles or videos that the user is interested in.

[0004] However, the above-mentioned personalized recommendation method is prone to the information cocoon problem. SUMMARY

[0005] The present disclosure provides a method and device for determining a resource to be recommended to solve the information cocoon problem and explore more resources that a user may be interested in.

[0006] According to a first aspect of the present disclosure, a method for determining a resource to be recommended is provided, comprising:

[0007] In the plurality of user vectors and the plurality of resource vectors, a user vector of a first user and a resource vector of a first resource are obtained, wherein the plurality of user vectors and the plurality of resource vectors are determined according to resource interaction behavior data of a plurality of users, and / or the plurality of user vectors and the plurality of resource vectors are determined according to random walk corpus of a user-resource relationship graph;

[0008] According to the similarity between the user vector of the first user and the resource vector of the first resource, it is determined whether to recall the first resource, and a first recalled resource is obtained;

[0009] According to the recalled resource, a resource to be recommended to the first user is determined, and the recalled resource includes the first recalled resource.

[0010] According to a second aspect of the present disclosure, a device for determining a resource to be recommended is provided, comprising:

[0011] The acquisition unit is configured to acquire a user vector of a first user and a resource vector of a first resource from a plurality of user vectors and a plurality of resource vectors, wherein the plurality of user vectors and the plurality of resource vectors are determined according to resource interaction behavior data of a plurality of users, and / or the plurality of user vectors and the plurality of resource vectors are determined according to random walk corpus of a user-resource relationship graph.

[0012] The first recall unit is configured to determine whether to recall the first resource according to a similarity between the user vector of the first user and the resource vector of the first resource, to obtain a first recalled resource.

[0013] The determination unit is configured to determine a to-be-recommended resource for the first user according to the recalled resource, wherein the recalled resource includes the first recalled resource.

[0014] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0015] at least one processor; and

[0016] a memory connected with the at least one processor in communication;

[0017] wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for determining the to-be-recommended resource according to the first aspect.

[0018] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to perform the method for determining the to-be-recommended resource according to the first aspect.

[0019] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising: a computer program stored in a readable storage medium, and at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to enable the electronic device to perform the method for determining the to-be-recommended resource according to the first aspect.

[0020] According to the technical solution provided in this disclosure, user vectors and resource vectors for resource recall can be obtained in multiple ways. One method is to determine multiple user vectors and resource vectors based on resource interaction behavior data of multiple users. Another method is to determine multiple user vectors and resource vectors based on random walk corpora of user-resource relationship graphs. This improves the diversity and completeness of the generation methods of user vectors and resource vectors, which is conducive to generating more and more accurate user vectors and resource vectors. The user vector of the first user is obtained from multiple user vectors, and the resource vector of the first resource is obtained from multiple resource vectors. Based on the similarity between the user vector of the first user and the resource vector of the first resource, it is determined whether to recall the first resource, thus obtaining the first recalled resource. This achieves vector-based resource recall. More and more accurate user vectors and resource vectors improve the diversity and accuracy of the first recalled resource, uncovering more first recalled resources and improving the accuracy of identifying the first recalled resource as a resource of interest to the first user. Based on the recalled resources containing the first recalled resource, resources to be recommended to the first user are determined, thereby more accurately exploring resources that the user may be interested in and recommending them to the user, solving the information cocoon problem.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0022] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0023] Figure 1 This is a schematic diagram illustrating an application scenario to which this disclosure applies;

[0024] Figure 2 This is a flowchart illustrating the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 ;

[0025] Figure 3 This is a flowchart illustrating the process of determining the first user vector and the first resource vector based on a random walk corpus of a user-resource relationship graph in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 ;

[0026] Figure 4 This is a flowchart illustrating the process of determining a second user vector and a second resource vector based on resource interaction behavior data of multiple users in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 ;

[0027] Figure 5This is a flowchart illustrating the process of determining a second user vector and a second resource vector based on resource interaction behavior data of multiple users in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 2 ;

[0028] Figure 6 This is an example diagram of a first recall pathway provided according to an embodiment of this disclosure;

[0029] Figure 7 This is a flowchart illustrating the method for determining resources to be recommended according to embodiments of this disclosure. Figure 2 ;

[0030] Figure 8 This is an example diagram of a second recall pathway provided according to an embodiment of this disclosure;

[0031] Figure 9 Schematic diagram of the structure of the apparatus for determining the resource to be recommended provided in the embodiments of this disclosure Figure 1 ;

[0032] Figure 10 Schematic diagram of the structure of the apparatus for determining the resource to be recommended provided in the embodiments of this disclosure Figure 2 ;

[0033] Figure 11 This is a schematic block diagram of an example electronic device 1100 that can be used to implement embodiments of the present disclosure. Detailed Implementation

[0034] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0035] First, some of the terms used in the embodiments of this disclosure will be explained:

[0036] Exploit and Explore: Exploit refers to using users' historical behavior information to make more reliable and accurate resource recommendations, such as recommending resources that the user has clicked more than 3 times or resources related to the user's points of interest that the user has clicked more than 3 times; Explore refers to making unreliable resource recommendations to explore resources that the user may be interested in.

[0037] Recall, Ranking, and Fusion: Recommendation algorithms can be divided into three stages: recall, ranking, and fusion. Recall refers to selecting user-relevant resources from a massive pool of resources as a candidate set for recommendation; ranking refers to sorting and truncating the candidate set, for example, selecting 100 resources from 10,000; fusion refers to selecting the final recommended resources from the sorted resources using ranking and recommendation rules (such as quantity limits for each type of resource).

[0038] Resource recall methods for exploring resources that users may be interested in: Recommendation algorithms used to explore resources that users may be interested in can include content-based recommendation and model-based recommendation. Content-based recommendation uses content such as points of interest and categories, which are highly interpretable, so it is also called explicit recall. Model-based recommendation, also known as embedding-based recommendation, is less interpretable than explicit recall, so it is also called implicit recall. For example, if a user likes dance videos, then dance can be considered a point of interest, and content related to this point of interest can be recalled for the user; this is explicit recall. Training a model using corpus samples related to resources, and then using the model to generate vectors corresponding to the resources, and recalling resources based on these vectors, is implicit recall because the vectors themselves are not interpretable.

[0039] In personalized recommendations, to avoid the information cocoon problem, a trade-off needs to be struck between utilization and exploration. This involves using user information for highly confident recommendations while also appropriately exploring the furthest possible interests the user might have, broadening their horizons. Simultaneously, the proportion of exploration must be controlled to prevent it from lowering the accuracy of the recommendations. Related technologies primarily focus on the utilization aspect of recommendations, specifically researching the use of user information for highly confident recommendations, thus exhibiting the information cocoon problem.

[0040] To address the aforementioned issues and uncover more hidden user interests—that is, to discover more resources that users might be interested in—this disclosure proposes a method, apparatus, device, and storage medium for determining resources to be recommended. Applied to the fields of big data, information flow, and intelligent recommendation in data processing technology, this method proposes to utilize multiple user vectors and multiple resource vectors obtained from resource interaction behavior data of multiple users and / or random walk corpora of user-resource relationship graphs to achieve resource recall. This is equivalent to proposing multiple resource recall methods: resource recall based on user vectors and resource vectors determined from user interaction behavior data, and / or, resource recall based on user vectors and resource vectors determined from random walk corpora of user-resource relationship graphs. Therefore, it improves the diversity and completeness of resource recall methods for exploring resources that users may be interested in, enhances the diversity and accuracy of recalled resources, and can more accurately explore more resources that users might be interested in, effectively solving the information cocoon problem.

[0041] Figure 1 This diagram illustrates an application scenario applicable to an embodiment of this disclosure, which can be a personalized recommendation scenario. In this application scenario, the devices involved include a resource recommendation device 101, a first recall device 102, and a resource database 103. The resource recommendation device 101 and the first recall device 102 can be servers or terminals. Figure 1 Taking the resource recommendation device 101 and the first recall device 102 as servers as an example; the resource database 103 stores multiple resources.

[0042] In a personalized recommendation scenario, the first recall device 102 can recall resources based on multiple user vectors and multiple resource vectors to determine the first recalled resource. If the resource recommendation device 101 determines that the first recalled resource is to be recommended, it can retrieve the corresponding resource from the resource database 103 for recommendation. The determination of the multiple user vectors and multiple resource vectors can be performed on the first recall device 102 or on other devices (not shown in the figure).

[0043] Optionally, the application scenario also includes a second recall device 104, which can recall resources based on the interests that users may be interested in, and determine the second recall resources. If the resource recommendation device 101 determines to recommend the second recall resources, it can obtain the corresponding resources from the resource database 103 for recommendation.

[0044] Optionally, the application scenario also includes a third recall device 105, which can recall resources based on confidence and determine the recalled resources to be used (rather than recommended for exploration). If the resource recommendation device 101 determines that the recalled resource is recommended, it can obtain the corresponding resource from the resource database 103 for recommendation.

[0045] The first recall device 102, the second recall device 104, the third recall device 105, and the resource recommendation device 101 can be the same device or different devices. Figure 1 Taking the first recall device 102, the second recall device 104, the third recall device 105, and the resource recommendation device 101 as different servers as an example, the first recall device 102, the second recall device 104, and the third recall device 105 can respectively communicate with the resource recommendation device 101 via wired or wireless communication. Figure 1 Take wireless as an example.

[0046] Optionally, the application scenario also includes terminal 106. Terminal 106 can send a resource recommendation request to resource recommendation device 101. Resource recommendation device 101 can respond to the request, determine the resources to be recommended, further filter the resources, and then send them to terminal 106. Alternatively, it can directly send the resources to be recommended to terminal 106 for resource recommendation. The resource recommendation device 101 and terminal 106 can communicate via wired or wireless communication. Figure 1 Take wireless as an example.

[0047] The technical solutions of this disclosure and how they solve the aforementioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this disclosure will now be described with reference to the accompanying drawings.

[0048] For example, the execution subject of this disclosure embodiment can be an electronic device, which can be a server or a terminal. The server can be a centralized server, a distributed server, or a cloud server. The terminal can be a personal digital assistant (PDA) device, a handheld device with wireless communication capabilities (e.g., a smartphone, tablet), a computing device (e.g., a personal computer, PC), an in-vehicle device, a wearable device (e.g., a smartwatch, smart bracelet), and a smart home device (e.g., a smart speaker, smart display device), etc.

[0049] Figure 2 This is a flowchart illustrating the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 .like Figure 2 As shown, the methods for determining the resources to be recommended include:

[0050] S201, among multiple user vectors and multiple resource vectors, obtain the user vector of the first user and the resource vector of the first resource, wherein the multiple user vectors and multiple resource vectors are determined based on the resource interaction behavior data of multiple users, and / or, the multiple user vectors and multiple resource vectors are determined based on the random walk corpus of the user-resource relationship graph.

[0051] In this context, among multiple user vectors and multiple resource vectors, one user corresponds to one user vector, and one resource corresponds to one resource vector. A user's user vector can be understood as the user's feature vector, and a resource's resource vector can be understood as the resource's feature vector.

[0052] User resource interaction behavior data can include user attribute information, attribute information of the resources the user interacts with, and user interaction behaviors with those resources. User attribute information includes user ID, age group, and online time; resource attribute information includes resource type (e.g., article, video, music), style (e.g., serious, humorous), author, and publication time; and user interaction behaviors include clicking, browsing, and downloading. Since user attribute information reflects user characteristics, resource attribute information reflects resource characteristics, and interaction behaviors reflect interaction characteristics, feature extraction can be performed on user resource interaction behavior data to determine the user vector and the resource vector of the interacting resources. Therefore, multiple user vectors and multiple resource vectors can be determined based on the resource interaction behavior data of multiple users.

[0053] The user-resource relationship graph contains multiple nodes and edges. The nodes include user nodes and resource nodes, and the edges connect them. When a user corresponding to a user node has interacted with a resource corresponding to a resource node, an edge exists between the user node and the resource node; that is, the edges in the user-resource relationship graph reflect user interaction behavior with resources. The random walk corpus of the user-resource relationship graph is obtained by performing weighted random walks on the graph. Since multiple user nodes in the graph reflect the user characteristics of multiple users, multiple resource nodes reflect the resource characteristics of multiple resources, and multiple edges reflect multiple interaction characteristics, the random walk corpus can also reflect user characteristics, resource characteristics, and interaction characteristics. Therefore, multiple user vectors and multiple resource vectors can be determined based on the random walk corpus.

[0054] In this context, the first user can be any user among multiple user vectors, from whom resources are to be recommended. The first resource can be any candidate resource selected from multiple resources corresponding to multiple resource vectors. For example, resources published within the last week can be selected as the first resource based on their publication time, or resources with a access count exceeding a certain threshold can be selected as the first resource based on their popularity. The first resource can be either a resource that the first user has interacted with or a resource that the first user has not interacted with, in order to achieve the purpose of resource exploration, rather than being limited to recommending resources that the first user has interacted with, thus avoiding the information cocoon problem.

[0055] In this embodiment, the user vector of the first user can be obtained from multiple user vectors based on the user identifier of the first user; the resource identifier of the first resource can be obtained from multiple resource vectors based on the resource identifier of the first resource. The number of first users and the number of first resources can be one or more.

[0056] In one possible implementation, upon receiving a resource recommendation request from a first user, the system can search for the first user's user vector among multiple user vectors based on the first user's user identifier. Then, based on resource filtering criteria, a first resource is determined from among multiple resources. Finally, the resource vector of the first resource is searched among multiple resource vectors based on its resource identifier. The first user's user identifier can be obtained from the resource recommendation request or determined based on the terminal identifier of the terminal sending the request. This allows for the timely determination of recommended resources for the first user requesting resource recommendations.

[0057] Furthermore, the system can retrieve the user vector of the first user from multiple user resources in response to the first user's online behavior; or, it can periodically retrieve the user vector of the first user from multiple user resources to achieve periodic resource recommendations. These will not be described in detail here.

[0058] S202, based on the similarity between the user vector of the first user and the resource vector of the first resource, determine whether to recall the first resource and obtain the first recalled resource.

[0059] In this embodiment, the user vector of a first user can be compared with the resource vector of a first resource to obtain the similarity between the user vector of the first user and the resource vector of the first resource. When there are multiple first resources, the similarity between the user vector of the first user and the resource vector of each first resource can be determined. The higher the similarity between the user vector of the first user and the resource vector of the first resource, the higher the degree of interest of the first user in the first resource. Therefore, if the similarity between the user vector of the first user and the resource vector of the first resource is greater than a threshold, it can be determined that the first resource should be recalled to obtain the first recalled resource; alternatively, the first resources can be sorted according to the similarity between the user vector of the first user and the resource vector of the first resource, and the first recalled resources can be determined to include the first resources ranked from high to low with a preset number of similarities.

[0060] The similarity between the user vector of the first user and the resource vector of the first resource can be the cosine similarity, Euclidean distance, Manhattan distance, etc., which will not be listed one by one here, nor will the specific calculation process of the similarity be described.

[0061] S203, Based on the recalled resources, determine the resources to be recommended to the first user, including the first recalled resources.

[0062] In this embodiment, after obtaining the first recalled resource, the recommendation weight of the first recalled resource can be determined based on the similarity between the user vector of the first user and the resource vector of the first recalled resource. For example, the recommendation weight of the first recalled resource can be determined as the similarity between the user vector of the first user and the resource vector of the first recalled resource, or the recommendation weight of the first recalled resource can be obtained by performing corresponding calculations on the similarity between the user vector of the first user and the resource vector of the first recalled resource. The higher the similarity, the greater the recommendation weight, and the greater the recommendation weight, the greater the probability of being recommended to the first user. After obtaining the recalled resource, resources to be recommended to the first user can be determined from the recalled resource based on the recommendation weight of the recalled resource.

[0063] In this embodiment, multiple user vectors and multiple resource vectors for resource recall are obtained based on resource interaction behavior data of multiple users and / or random walk corpora of user-resource relationship graphs. During the recall process, a first recall resource is determined based on the similarity between the user vector of the first user and the resource vector of the first resource. Then, resources to be recommended to the first user are determined from the recall resources containing the first recall resource. Thus, by determining the user vectors and resource vectors for resource recall through one or more methods, the diversity and completeness of the methods for determining user vectors and resource vectors are improved, as are the diversity and accuracy of user vectors and resource vectors. This enables resource recall based on user vectors and resource vectors to explore and obtain more accurate resources, recommending resources obtained through exploration to users and solving the information cocoon problem.

[0064] It should be noted that the process of determining user vectors and resource vectors can be performed on the same device as the method for determining the resources to be recommended, or it can be performed on other devices. For example, user vectors and resource vectors can be determined on one server, while the resources to be recommended can be determined on another server.

[0065] Since multiple user vectors and multiple resource vectors can be determined either based on resource interaction behavior data of multiple users or based on random walk corpus of user-resource relationship graphs, the following provides an embodiment of the determination process of the first user vector and the first resource vector, and an embodiment of the determination process of the second user vector and the second resource vector, taking the determination of the first user vector and the first resource vector among multiple user vectors based on random walk corpus of user-resource relationship graphs, and the determination of the second user vector and the second resource vector among multiple resource vectors based on resource interaction behavior data of multiple users as examples.

[0066] The first user vector can be multiple user vectors or a subset of multiple user vectors; the first resource vector can be multiple resource vectors or a subset of multiple resource vectors. Similarly, the second user vector can be multiple user vectors or a subset of multiple user vectors; the second resource vector can be multiple resource vectors or a subset of multiple resource vectors.

[0067] For example, when determining multiple user vectors and multiple resource vectors solely based on resource interaction behavior data from multiple users, the first user vector is composed of multiple user vectors, and the first resource vector is composed of multiple resource vectors. When determining multiple user vectors and multiple resource vectors solely based on random walk corpora of user-resource relationship graphs, the second user vector is composed of multiple user vectors, and the second resource vector is composed of multiple resource vectors. When determining multiple user vectors and multiple resource vectors by combining resource interaction behavior data from multiple users and random walk corpora of user-resource relationship graphs, both the first and second user vectors can be partial user vectors from multiple user vectors, and both the first and second resource vectors can be partial resource vectors from multiple resource vectors. In this case, the first and second user vectors may include user vectors belonging to the same user. For example, after determining a user's user vector based on their resource interaction behavior, the user vector is then determined again based on the random walk corpora of the user-resource relationship graph (equivalent to updating the user's user vector). Similarly, the first and second resource vectors may include resource vectors belonging to the same resource; examples are not provided here.

[0068] Figure 3 This is a flowchart illustrating the process of determining the first user vector and the first resource vector based on a random walk corpus of a user-resource relationship graph in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 .like Figure 3 As shown, the process of determining the first user vector and the first resource vector based on the random walk corpus of the user-resource relationship graph includes:

[0069] S301, Determine the user resource relationship diagram.

[0070] In this embodiment, a user resource relationship graph can be obtained from a database; alternatively, a user resource relationship graph can be constructed based on resource interaction behavior data of multiple users; or alternatively, the user resource relationship graph can be updated based on newly added resource interaction behavior data of one or more users.

[0071] In this embodiment, during the process of constructing the user resource relationship graph, user nodes corresponding to multiple users can be constructed based on the attribute information of multiple users in the resource interaction behavior data of multiple users; resource nodes corresponding to multiple resource distributions can be constructed based on the attribute information of multiple resources in the resource interaction behavior data of multiple users; and edges connecting the corresponding user nodes and the corresponding resource nodes can be constructed based on the interaction behavior between users and resources.

[0072] In this embodiment, during the process of updating the user resource relationship graph, it can be determined whether a user node corresponding to the user exists in the user resource relationship graph based on the user's newly added resource interaction behavior data. If not, the user node corresponding to the user is constructed based on the user's attribute information in the user's newly added resource interaction behavior data. It can also be determined whether a resource node corresponding to the resource in the user's newly added resource interaction behavior data exists in the user resource relationship graph. If not, the resource node corresponding to the resource is constructed based on the resource's attribute information in the user's newly added resource interaction behavior data. Finally, the edges between the user node corresponding to the user and the resource node corresponding to the resource in the user's newly added resource interaction behavior data are updated based on the user's newly added resource interaction behavior data.

[0073] In one possible implementation, the weight of the edge connecting the user node and the resource node in the user-resource relationship graph can be determined based on the number of interactions between the user corresponding to the user node and the resource corresponding to the resource node. Thus, the edge weights reflect the user's interaction with the resource, improving the accuracy of the random walk corpus obtained through weighted random walks and enhancing the accuracy of resource exploration.

[0074] Furthermore, S301 includes: acquiring resource interaction behavior data of the second user; determining the number of interactions between the second user and the second resource based on the resource interaction behavior data of the second user, wherein the resource interaction behavior data of the second user includes the interaction behavior data of the second user with the second resource; in the user-resource relationship graph, determining the user node corresponding to the second user and the resource node corresponding to the second resource, and determining the weight of the edge between the user node corresponding to the second user and the resource node corresponding to the second resource based on the reciprocal of the number of interactions. Thus, by determining the edge weight based on the reciprocal of the number of interactions, the random walk corpus of the user-resource relationship graph can include resources that are relevant to the user but have relatively low confidence, achieving the purpose of resource exploration.

[0075] The second user may or may not include the first user.

[0076] The resource interaction behavior data of the second user can be newly added resource interaction behavior data of the second user, or it can be the resource interaction data used when initially constructing the user resource relationship graph.

[0077] In this implementation, if the weight of the edge connecting a user node and a resource node in the user resource relationship graph is determined by the number of interactions between the user corresponding to the user node and the resource corresponding to the resource node, then the edge weight will be higher for more interactions. During a weighted random walk, the walk will reach the resource node corresponding to the resource with more user interactions, resulting in resources with higher confidence. This achieves resource recommendation based on utilization, rather than resource recommendation based on exploration. To improve the exploration effect, this implementation determines the weight of the edge connecting the user node corresponding to the second user and the resource node corresponding to the second resource based on the reciprocal of the number of interactions between the second user and the second resource. This ensures that the fewer interactions the second user has with the second resource, the higher the weight of the edge connecting the user node corresponding to the second user and the resource node corresponding to the second resource. This allows the weighted random walk to reach resources with fewer interactions with the user. Thus, by performing a weighted random walk in the user resource relationship graph, more resources that the user might be interested in can be explored.

[0078] Furthermore, the weight of the edge connecting the user node corresponding to the second user and the resource node corresponding to the second resource can be determined as the reciprocal of the number of interactions between the second user and the second resource. This makes the weight of the edge directly related to the number of interactions between the user and the resource, while ensuring that the larger the number of interactions, the smaller the weight of the edge. This allows the weighted random walk in the user-resource relationship graph to reach resources with fewer interactions with the user, thus improving the resource exploration effect.

[0079] S302, perform a weighted random walk in the user resource relationship graph to obtain the random walk corpus.

[0080] In this embodiment, a random walk algorithm can be used to perform a weighted random walk in the user-resource relationship graph based on the edge weights, resulting in a random walk path. The random walk path includes user nodes, resource nodes, and edges along the path. Based on the attribute information of the users corresponding to the user nodes, the attribute information of the resources corresponding to the resource nodes, and the edge weights, a random walk corpus can be obtained. This random walk corpus may include attribute information of one or more users, attribute information of one or more resources, and the weights of one or more edges.

[0081] For example: First, determine the starting point of the random walk as user node a1. User node a1 connects to resource nodes b1 and b2, indicating that the user corresponding to user node a1 has interacted with the resources corresponding to resource nodes b1 and b2. Based on the weights of the edges between user node a1 and resource node b1, and the weights of the edges between user node a1 and resource node b2, the next node is determined through the random walk algorithm, for example, the next node is resource node b1. In this way, the walk is performed step by step. The walk can be stopped when the number of walked nodes exceeds the threshold or when there is no next node after the walked node, thus obtaining the random walk corpus.

[0082] There are no restrictions on the random walk algorithm here.

[0083] S303, Encode the random walk corpus to obtain the first encoding vector.

[0084] In this embodiment, one or more random walk corpora can be obtained from the user resource relationship graph. For each random walk corpus, the random walk corpus can be encoded by an encoding algorithm to obtain the first encoding vector corresponding to the random walk corpus.

[0085] In one possible implementation, a skip-gram model can be used to encode the random walk corpus, obtaining the first encoding vector corresponding to the random walk corpus, thereby improving the encoding accuracy of the random walk corpus through the skip-gram model.

[0086] S304, the first encoding vector is input into the first neural network, and user features and resource features are extracted from the first encoding vector in the first neural network to obtain the first user vector among multiple user vectors and the first resource vector among multiple resource vectors.

[0087] In this embodiment, the first neural network may include a feature processing network for extracting user features and a feature processing network for extracting resource features. A first encoded vector can be input into both the feature processing network for extracting user features and the feature processing network for extracting resource features. In the feature processing network for extracting user features, user features are extracted from the first encoded vector to obtain a first user vector; in the feature processing network for extracting resource features, resources are extracted from the first encoded vector to obtain a first resource vector. Since the weighted random walk may traverse one or more user nodes and one or more resource nodes, one or more first user vectors and one or more first resource vectors can be obtained based on a single random walk corpus.

[0088] In this embodiment, a weighted random walk is performed on the user-resource relationship graph to obtain a random walk corpus. This corpus is then encoded to obtain a first encoded vector. A first neural network is used to extract user and resource features from the first encoded vector, resulting in a first user vector and a first resource vector. This combination of the random walk corpus and the neural network improves the accuracy of user and resource vector generation. In particular, determining the edge weights in the user-resource relationship graph based on the reciprocal of the number of user-resource interactions enhances the accuracy of identifying resources that a user might be interested in, thereby improving resource exploration effectiveness.

[0089] In some embodiments, such as Figure 3 As shown, the process of determining the first user vector and the first resource vector based on the random walk prediction of the user-resource relationship graph also includes:

[0090] S305, determine the first similarity between the first user vector and the first resource vector.

[0091] In this embodiment, the first user vector can be matched with the first resource vector to obtain the similarity between the first user vector and the first resource vector, i.e., the first similarity.

[0092] In one possible implementation, the first similarity can be determined as the cosine similarity between the first user vector and the first resource vector.

[0093] In another possible implementation, the first neural network includes a vector matching layer for determining the similarity between vectors. A first user vector and a first resource vector are input to the vector matching layer, where the first user vector and the first resource vector are matched to obtain a first similarity. Thus, the accuracy of the first similarity is improved through the first neural network.

[0094] S306, determine the first error value based on the actual interaction behavior of the user corresponding to the first user vector with the resource corresponding to the first resource vector and the first similarity.

[0095] Here, the first similarity reflects the predicted probability that the user corresponding to the first user vector will interact with the resource corresponding to the first resource vector; the higher the first similarity, the higher the predicted probability. The actual interaction behavior of the user corresponding to the first user vector with the resource corresponding to the first resource vector reflects whether the user corresponding to the first user vector has interacted with the resource corresponding to the first resource vector.

[0096] In this embodiment, if the user corresponding to the first user vector actually interacts with the resource corresponding to the first resource vector, the actual interaction label of the user corresponding to the first user vector with the resource corresponding to the first resource vector can be determined as a first value (e.g., 1). If the user corresponding to the first user vector does not actually interact with the resource corresponding to the first resource vector, the actual interaction label of the user corresponding to the first user vector with the resource corresponding to the first resource vector can be determined as a second value (e.g., 0). The first similarity can be compared with the actual interaction label to obtain a first error value.

[0097] S307, Adjust the parameters of the first neural network according to the first error value.

[0098] In this embodiment, the parameters of the first neural network can be adjusted according to the first error value and the model optimization algorithm (such as the gradient optimization algorithm, which is not limited here) to achieve model training of the first neural network.

[0099] Therefore, in the process of generating the first user vector and the first resource vector, the first neural network is optimized based on the similarity between the first user vector and the first resource vector, effectively improving the accuracy of the first neural network in generating user vectors and resource vectors. Furthermore, since the first neural network can also output vector similarity, the accuracy of its predicted vector similarity can also be improved.

[0100] In some embodiments, the user-resource relationship graph can be a user-resource bipartite graph, which can clearly and concisely describe the interaction relationship between users and resources.

[0101] Figure 4 This is a flowchart illustrating the process of determining a second user vector and a second resource vector based on resource interaction behavior data of multiple users in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 1 .like Figure 4 As shown, the process of determining the second user vector and the second resource vector based on resource interaction behavior data of multiple users includes:

[0102] S401, encode the resource interaction behavior data of multiple users to obtain the second encoding vector.

[0103] Different users can have different resource interaction behavior data.

[0104] In this embodiment, for each user's resource interaction behavior data, an encoding algorithm can be used to encode the user's resource interaction behavior data to obtain a second encoding vector corresponding to the user's resource interaction behavior data. Therefore, multiple second encoding vectors can be obtained.

[0105] In one possible implementation, a one-hot encoding model can be used to encode the user's resource interaction behavior data, resulting in a second encoding vector corresponding to the user's resource interaction behavior data. This improves the encoding accuracy of the resource interaction behavior data through the one-hot encoding model.

[0106] S402, the second encoding vector is input into the second neural network, and user features and resource features are extracted from the second encoding vector in the second neural network to obtain the second user vector from multiple user vectors and the second resource vector from multiple resource vectors.

[0107] The network parameters of the second neural network are related to the similarity between the third user vector generated by the third neural network and the third resource vector generated by the third neural network. The greater the similarity between the third user vector and the third resource vector, the smaller the network parameters of the second neural network. The second resource vector, third resource vector, second user vector, and third user vector are only used to distinguish whether the resource vector is output by the second neural network or the third neural network.

[0108] The third neural network can be a pre-trained neural network used to predict the similarity between the third user vector and the third resource vector. The more interactions the user corresponding to the third user vector has with the resource corresponding to the third resource vector, the higher the similarity predicted by the third neural network. Considering that the purpose of this embodiment is to explore resources that the user may be interested in, these resources are those that the user has not interacted with or has interacted with very few times. In order for the second neural network to explore such resources, that is, to explore resource vectors that are similar to the user vector but not so similar, the network parameters of the second neural network can be determined based on the similarity between the third user vector and the third resource vector output by the third neural network, so that the greater the similarity between the third user vector and the third resource vector, the smaller the network parameters.

[0109] In this embodiment, the second neural network may include a feature processing network for extracting user features and a feature processing network for extracting resource features. The second encoded vector can be input into both the feature processing network for extracting user features and the feature processing network for extracting resource features. In the feature processing network for extracting user features, user features are extracted from the second encoded vector to obtain a second user vector; in the feature processing network for extracting resource features, resources are extracted from the second encoded vector to obtain a second resource vector.

[0110] In this embodiment, the similarity between the third user vector and the third resource vector output by the third neural network follows the rule that the more times the user corresponding to the third user vector interacts with the resource corresponding to the third resource vector, the higher the similarity. The network parameters of the second neural network are determined based on the similarity between the third user vector and the third resource vector, such that the network parameters of the second neural network are negatively correlated with the similarity between the third user vector and the third resource vector; that is, the higher the similarity, the smaller the network parameters of the second neural network. This allows the second user vector and the second resource vector output by the second neural network to explore resources that users have not interacted with or have interacted with infrequently, improving resource exploration effectiveness and effectively solving the information cocoon problem.

[0111] In some embodiments, such as Figure 4 As shown, the process of determining the second user vector and the second resource vector based on resource interaction behavior data of multiple users also includes:

[0112] S403, determine the second similarity between the second user vector and the second resource vector.

[0113] In this embodiment, the second user vector can be matched with the second resource vector to obtain the similarity between the second user vector and the second resource vector, i.e., the second similarity.

[0114] In one possible implementation, the second similarity can be determined as the cosine similarity between the second user vector and the second resource vector.

[0115] In another possible implementation, the second neural network includes a vector matching layer for determining the similarity between vectors. The second user vector and the second resource vector are input into the vector matching layer, where they are matched to obtain a second similarity score. Thus, the accuracy of the second similarity score is improved through the second neural network.

[0116] S404, determine the second error value based on the actual interaction behavior of the user corresponding to the second user vector with the resource corresponding to the second resource vector and the second similarity.

[0117] The second similarity reflects the predicted probability that the user corresponding to the second user vector will interact with the resource corresponding to the second resource vector. The higher the second similarity, the higher the predicted probability.

[0118] In this embodiment, the actual interaction behavior of the user corresponding to the second user vector with the resource corresponding to the second resource vector can be reflected as the actual interaction label of the user corresponding to the second user vector with the resource corresponding to the second resource vector. Therefore, the second similarity can be compared with this actual interaction label to obtain the second error value.

[0119] S405, Adjust the parameters of the second neural network based on the second error value.

[0120] In this embodiment, the parameters of the second neural network can be adjusted according to the second error value and the model optimization algorithm to achieve model training of the second neural network.

[0121] Therefore, in the process of generating the second user vector and the second resource vector, the second neural network is optimized based on the similarity between the second user vector and the second resource vector, effectively improving the accuracy of the second neural network in generating user vectors and resource vectors. Furthermore, since the second neural network can also output vector similarity, the accuracy of its predicted vector similarity can also be improved.

[0122] Figure 5 This is a flowchart illustrating the process of determining a second user vector and a second resource vector based on resource interaction behavior data of multiple users in the method for determining resources to be recommended according to embodiments of this disclosure. Figure 2 .like Figure 5 As shown, the process of determining the second user vector and the second resource vector based on resource interaction behavior data of multiple users includes:

[0123] S501, encode the resource interaction behavior data of multiple users to obtain the second encoding vector.

[0124] The implementation principle and technical effects of S501 can be referred to in the aforementioned embodiments, and will not be repeated here.

[0125] S502, the second encoding vector is input into the third neural network. In the third neural network, user features and resource features are extracted from the second encoding vector to obtain the third user vector and the third resource vector.

[0126] In this embodiment, the third neural network may include a feature processing network for extracting user features and a feature processing network for extracting resource features. The second encoded vector can be input into both the feature processing network for extracting user features and the feature processing network for extracting resource features. In the feature processing network for extracting user features, user features are extracted from the second encoded vector to obtain a third user vector; in the feature processing network for extracting resource features, resources are extracted from the second encoded vector to obtain a third resource vector.

[0127] S503, update the network parameters of the second neural network based on the reciprocal of the similarity between the third user vector and the three resource vectors.

[0128] In this embodiment, the similarity between the third user vector and the third resource vector can be determined. The third neural network may include a vector matching layer for determining the similarity between vectors. The third user vector and the third resource vector are input into the vector matching layer, where the third user vector and the third resource vector are matched to obtain the similarity between the third user vector and the third resource vector. This improves the accuracy of the similarity between the third user vector and the third resource vector through the third neural network. Alternatively, the similarity between the third user vector and the third resource vector can be the cosine similarity between the three user vectors. Then, the network parameters of the second neural network can be updated based on the reciprocal of the similarity between the third user vector and the three resource vectors, such that the greater the similarity between the third user vector and the three resource vectors, the smaller the network parameters of the second neural network. Therefore, in each process of using the second neural network to determine the user vector and resource vector, the encoded vector obtained from the same resource interaction behavior data by the first neural network is processed to obtain a similarity that is positively correlated with the number of interactions. The network parameters of the second neural network are determined based on the reciprocal of this similarity. As a result, the similarity between the second user vector and the second resource vector output by the second neural network is not affected by the fact that the greater the number of interactions, the higher the similarity, and may even be negatively correlated with the number of interactions, thereby improving the accuracy of resource exploration based on the second user vector and the second resource vector.

[0129] In one possible implementation, S503 includes updating the network parameters of the random sampling layer in the second neural network to the inverse of the similarity between the third user vector and the third resource vector.

[0130] In this implementation, the feature processing network of the second neural network used to extract resource features includes a random sampling layer. The network parameters of this random sampling layer can be updated to the inverse of the similarity between the third user vector and the third resource vector. Thus, during the sampling of resource features in the random sampling layer, more resource features of resources with fewer user interactions or no interaction can be collected. This allows for the exploration of resources with fewer user interactions or no interaction during resource retrieval, thereby improving the resource exploration effect.

[0131] S504, the second encoding vector is input into the second neural network, and user features and resource features are extracted from the second encoding vector in the second neural network to obtain the second user vector from multiple user vectors and the second resource vector from multiple resource vectors.

[0132] The implementation principle and technical effects of S504 can be referred to in the aforementioned embodiments, and will not be repeated here.

[0133] In some embodiments, such as Figure 5As shown, the process of determining the second user vector and the second resource vector based on resource interaction behavior data of multiple users also includes:

[0134] S505, determine the second similarity between the second user vector and the second resource vector.

[0135] S506, determine the second error value based on the actual interaction behavior of the user corresponding to the second user vector with the resource corresponding to the second resource vector and the second similarity.

[0136] S507, Adjust the parameters of the second neural network based on the second error value.

[0137] The implementation principles and technical effects of S505 to S507 can be referred to in the aforementioned embodiments, and will not be repeated here.

[0138] In this embodiment, during each process of determining user vectors and resource vectors using the second neural network, the network parameters of the second neural network are determined using the first neural network. Then, the second neural network generates the second user vector and the second resource vector. Furthermore, the second neural network can be trained based on the similarity between the second user vector and the second resource vector. This improves the accuracy of the user vectors and resource vectors generated by the second neural network in multiple ways, thereby enhancing the accuracy of resource exploration based on the user vectors and resource vectors generated by the second neural network.

[0139] In some embodiments, the first neural network, the second neural network, and the third neural network can all employ a Matchnet network. In a Matchnet network, the feature processing network for extracting user features and the feature processing network for extracting resource features are separate, which can improve the extraction effect of user features and resource features. Furthermore, the Matchnet network also includes a vector matching layer, which can obtain the similarity between user vectors and resource vectors, thereby improving the accuracy of similarity.

[0140] As an example, Figure 6 This is an example diagram of a first recall path provided in an embodiment of the present disclosure, wherein the first recall path can obtain a first recall resource set, which contains multiple first recall resources. For example... Figure 6As shown, the first recall pathway may include two recall queues (i.e., recall methods): one recall queue is an exploratory graph neural network (i.e., the first neural network), specifically, by constructing a user-resource bipartite graph, performing weighted random walks on the user-resource bipartite graph to obtain random walk corpus, then using the random walk corpus to generate user vectors and resource vectors on the graph neural network, and using the random walk corpus to train the graph neural network to obtain user vectors and resource vectors for resource recall; the other recall queue is an exploratory surprise factor model (i.e., the second neural network), in which the network parameters of the random sampling layer are determined by the similarity generated by the dual-tower correlation model (i.e., the third neural network). The random sampling layer is located in the feature processing network of the surprise factor model for extracting resource features, and this feature processing network also includes one or more intermediate layers for feature extraction. Figure 6 Taking two intermediate layers as an example, the similarity between user vectors and resource vectors is calculated. This similarity can be further used to adjust the model parameters of the surprise factor model. Figure 6 (This step is not shown), thus obtaining the user vector and resource vector for resource retrieval. It can be seen that the user vector and resource vector can be periodically updated through one or two of these methods to improve their accuracy; during resource recommendation, the user vector and resource vector can be used to achieve resource retrieval.

[0141] In some embodiments, the recalled resources also include second recalled resources recalled based on the potential interests of the first user, i.e., resources obtained through explicit recall. Based on this, determining the resources to be recommended to the first user according to the recalled resources may include: increasing the recommendation weights corresponding to the first and second recalled resources during resource sorting and / or resource fusion; and determining the resources to be recommended to the first user from the recalled resources based on the recommendation weights corresponding to the recalled resources. Thus, by increasing the weights corresponding to the first and second recalled resources respectively, the probability of the resources to be recommended containing both the first and second recalled resources is increased, thereby recommending the explored resources to the user and solving the information cocoon problem.

[0142] In some embodiments, during the process of determining the resources to be recommended to the first user from the recalled resources based on the recommendation weights corresponding to the recalled resources, the resources to be recommended to the first user can be determined from the recalled resources based on the proportion threshold corresponding to the exploratory resources and the recommendation weights corresponding to the recalled resources. This controls the proportion of the first and second recalled resources, preventing the exploratory resources from having an excessively high proportion in the resources to be recommended, which would lead to lower accuracy of the recommended resources.

[0143] In some embodiments, the recalled resources may also include resources recalled by confidence (i.e. by recommended use), and the recall process for such resources will not be described in detail here.

[0144] Below, an embodiment of the recall process for the second recall resource is provided.

[0145] Figure 7 This is a flowchart illustrating the method for determining resources to be recommended according to embodiments of this disclosure. Figure 2 .like Figure 7 As shown, the methods for determining the resources to be recommended include:

[0146] S701, among multiple user vectors and multiple resource vectors, obtain the user vector of the first user and the resource vector of the first resource, wherein the multiple user vectors and multiple resource vectors are determined based on the resource interaction behavior data of multiple users, and / or, the multiple user vectors and multiple resource vectors are determined based on the random walk corpus of the user-resource relationship graph;

[0147] S702, based on the similarity between the user vector of the first user and the resource vector of the first resource, determine whether to recall the first resource and obtain the first recalled resource.

[0148] The implementation principles and technical effects of S701 to S702 can be referred to in the aforementioned embodiments, and will not be repeated here.

[0149] S703 identifies the potential interests of the first user.

[0150] Among them, potential points of interest refer to the points of interest that the first user may be interested in.

[0151] In this embodiment, the potential interests of the first user can be determined through a non-confidence method in order to explore the interests that the user may be interested in.

[0152] In some embodiments, S703 includes at least one of the following implementations:

[0153] Method 1: Identify potential points of interest (POIs) as those whose click count meets the first user's requirement and whose most recent click time meets the time requirement. Since POIs are only those the first user might be interested in, not necessarily those they are certain to be interested in, the click and time requirements can be set lower. This lower requirement allows for the identification of POIs more accurate.

[0154] In Method 1, the user's interest-point interaction behavior data can be obtained. This data may include the user's attribute information, the attribute information of the interest points the user has interacted with, and the user's interaction behavior with those interest points. The number of clicks and the click time for each interest point can be obtained from this data. If the number of clicks meets a requirement and the click time meets a requirement, then that interest point is identified as a potential interest point.

[0155] As an example, a potential point of interest is identified as one that has been clicked by the first user twice and has been clicked by the first user within the last 45 days.

[0156] Method 2: Based on the similarity between the user vector of the first user and the interest point vectors of multiple popular interest points, potential interest points are identified among the multiple popular interest points. The interest point vectors are determined based on the interest point interaction data of multiple users. Therefore, by mining potential interest points from popular interest points, more potential interest points can be discovered, and the accuracy of potential interest points can be improved.

[0157] In Method Two, besides mining potential interest points from those interacted with by the first user, potential interest points can also be mined from those not interacted with by the first user. Considering that popular interest points have a higher probability of being interacted with, potential interest points can be mined from popular interest points. Popular interest points refer to interest points with high interaction frequency (e.g., a large number of people click on the interest point). After obtaining popular interest points, interest point vectors can be obtained. Based on the similarity between the first user's user vector and the interest point vectors of popular interest points, potential interest points are determined from among the popular interest points. For example, if the similarity between the first user's user vector and the interest point vector of a popular interest point is greater than a threshold, then that popular interest point is identified as a potential interest point.

[0158] Furthermore, the interest vectors of popular points of interest are determined based on the interest interaction behavior data of third-party users, who are users who have interacted with popular points of interest. For example, popular points of interest can be recommended to a group of users to test their interest in them, resulting in a corpus for generating interest vectors for popular points of interest, i.e., the interest interaction behavior data of third-party users. Specifically, this third-party interest interaction behavior data can be encoded to obtain a third encoded vector. This third encoded vector is then input into a fourth neural network, where user feature extraction and interest feature extraction are performed to obtain the user vector of the third user and the interest vectors of popular points of interest. In this way, interest vectors of multiple popular points of interest can be obtained for the purpose of mining potential points of interest.

[0159] In this process, after obtaining the user vector of the third user and the interest point vector of the popular interest points, the similarity between the user vector of the third user and the interest point vector of the popular interest points can be determined. Based on this similarity and the actual interaction behavior of the third user with the popular interest points, a third error value is determined. The parameters of the fourth neural network are adjusted based on the third error value, thereby improving the accuracy of the interest point vector generated by the fourth neural network and thus improving the accuracy of potential interest point mining.

[0160] Therefore, we can first recommend popular interest points to a group of users to obtain interest point vectors for training the fourth neural network and generating popular interest points. Then, for the remaining users, we can mine the users' potential interest points by matching the user vectors with the interest point vectors.

[0161] Furthermore, the fourth neural network can employ a dual-tower network.

[0162] Method 3: Identify potential points of interest as those related to the first user's search resources. For example, when a user performs a keyword search in the search box and obtains search results, the sidebar of the search page often displays some points of interest related to the search results. These points of interest related to the search structure can be identified as potential points of interest to uncover more potential points of interest.

[0163] Method four involves identifying potential points of interest from third-party data platforms. For example, if the recommendation system uses a search engine, it can obtain users' interests on video websites and identify those interests as potential points of interest on the search engine. This allows for the full utilization of interest points provided by third-party data platforms to uncover more potential points of interest.

[0164] S704, based on potential points of interest, recall resources to obtain a second recalled resource.

[0165] In this embodiment, the second recall resource can be determined as the resource corresponding to the potential point of interest based on the correspondence between points of interest and resources. One point of interest can correspond to one or more resources, and the resources corresponding to the point of interest contain content related to that point of interest.

[0166] S705, during the process of sorting and / or merging the recalled resources, the recommendation weight corresponding to the first recalled resource and the recommendation weight corresponding to the second recalled resource are increased.

[0167] S706. Based on the recommendation weight corresponding to the recalled resources, determine the resources to be recommended to the first user from the recalled resources.

[0168] The implementation principles and technical effects of S705 to S706 can be referred to in the aforementioned embodiments, and will not be repeated here.

[0169] In this embodiment, resource retrieval is achieved by mining potential points of interest (POIs). Furthermore, by providing multiple methods for mining POIs, various resource retrieval methods are implemented, improving the diversity and completeness of POI-based resource retrieval methods, thereby enhancing the diversity and accuracy of the second-retrieved resources. In addition, by combining POI-based resource retrieval with vector-based (user vector, resource vector) resource retrieval, the diversity and accuracy of resources retrieved through exploration are improved, effectively solving the information cocoon problem.

[0170] As an example, Figure 8 This is an example diagram of a second recall path provided according to an embodiment of the present disclosure. The second recall path can obtain a second recall resource set, which contains multiple second recall resources. For example... Figure 8 As shown, the second recall path may include four recall queues (i.e., recall methods): In the first recall queue (i.e., method one in the aforementioned embodiment), potential interest points are mined from the interest points visited by the user, and then resources are recalled according to the inverted zipper method (i.e., according to the correspondence between interest points and resources, one interest point is associated with one or more resources in the inverted zipper); In the second recall queue (i.e., method two in the aforementioned embodiment), potential interest points are mined from popular interest points through a dual-tower network, and then resources are recalled according to the inverted zipper method; In the third recall queue (i.e., method three in the aforementioned embodiment), potential interest points are mined by searching the related terms in the right sidebar, and then resources are recalled according to the inverted zipper method; In the fourth recall queue (i.e., method four in the aforementioned embodiment), potential interest points are mined through the interest points of a third-party data platform, and then resources are recalled according to the inverted zipper method.

[0171] Figure 9 Schematic diagram of the structure of the apparatus for determining the resource to be recommended provided in the embodiments of this disclosure Figure 1 .like Figure 9 As shown, the device 900 for determining the resources to be recommended includes:

[0172] The acquisition unit 901 is used to acquire the user vector of the first user and the resource vector of the first resource from multiple user vectors and multiple resource vectors, wherein the multiple user vectors and multiple resource vectors are determined based on the resource interaction behavior data of multiple users, and / or the multiple user vectors and multiple resource vectors are determined based on the random walk corpus of the user-resource relationship graph.

[0173] The first recall unit 902 is used to determine whether to recall the first resource based on the similarity between the user vector of the first user and the resource vector of the first resource, so as to obtain the first recalled resource.

[0174] The determining unit 903 is used to determine the resources to be recommended to the first user based on the recalled resources, wherein the recalled resources include the first recalled resources.

[0175] Figure 10 Schematic diagram of the structure of the apparatus for determining the resource to be recommended provided in the embodiments of this disclosure Figure 2 .like Figure 10 As shown, the device 1000 for determining the resources to be recommended includes:

[0176] The acquisition unit 1001 is used to acquire the user vector of the first user and the resource vector of the first resource from multiple user vectors and multiple resource vectors, wherein the multiple user vectors and multiple resource vectors are determined based on the resource interaction behavior data of multiple users, and / or the multiple user vectors and multiple resource vectors are determined based on the random walk corpus of the user-resource relationship graph.

[0177] The first recall unit 1002 is used to determine whether to recall the first resource based on the similarity between the user vector of the first user and the resource vector of the first resource, so as to obtain the first recalled resource.

[0178] The determining unit 1003 is used to determine the resources to be recommended to the first user based on the recalled resources, wherein the recalled resources include the first recalled resources.

[0179] In some embodiments, multiple user vectors and multiple resource vectors are obtained by: determining a user-resource relationship graph; performing a random walk in the user-resource relationship graph to obtain a random walk corpus; encoding the random walk corpus to obtain a first encoded vector; inputting the first encoded vector into a first neural network, and extracting user features and resource features from the first encoded vector in the first neural network to obtain a first user vector among multiple user vectors and a first resource vector among multiple resource vectors.

[0180] In some embodiments, determining a user resource relationship graph includes: acquiring resource interaction behavior data of a second user; determining the number of interactions between the second user and the second resource based on the resource interaction behavior data of the second user, wherein the resource interaction behavior data of the second user includes the interaction behavior data of the second user with the second resource; in the user resource relationship graph, determining the user node corresponding to the second user and the resource node corresponding to the second resource, and determining the weight of the edge between the user node corresponding to the second user and the resource node corresponding to the second resource based on the reciprocal of the number of interactions.

[0181] In some embodiments, after inputting the first encoded vector into a first neural network, extracting user features and resource features from the first encoded vector in the first neural network to obtain a first user vector from multiple user vectors and a first resource vector from multiple resource vectors, the method further includes:

[0182] Determine the first similarity between the first user vector and the first resource vector; determine the first error value based on the actual interaction behavior of the user corresponding to the first user vector with the resource corresponding to the first resource vector and the first similarity; adjust the parameters of the first neural network based on the first error value.

[0183] In some embodiments, the multiple user vectors and multiple resource vectors are obtained by: encoding the resource interaction behavior data of multiple users to obtain a second encoded vector; inputting the second encoded vector into a second neural network, and extracting user features and resource features from the second encoded vector in the second neural network to obtain a second user vector from the multiple user vectors and a second resource vector from the multiple resource vectors; wherein, the network parameters of the second neural network are related to the similarity between the third user vector generated by the third neural network and the third resource vector generated by the third neural network, and the greater the similarity between the third user vector and the third resource vector, the smaller the network parameters of the second neural network.

[0184] In some embodiments, before inputting the second encoded vector into the second neural network, the method further includes:

[0185] The second encoded vector is input into the third neural network. In the third neural network, user features and resource features are extracted from the second encoded vector to obtain the third user vector and the third resource vector. The network parameters of the second neural network are updated based on the reciprocal of the similarity between the third user vector and the three resource vectors.

[0186] In some embodiments, updating the network parameters of the second neural network based on the reciprocal of the similarity between the third user vector and the three resource vectors includes: updating the network parameters of the random sampling layer in the second neural network to the reciprocal of the similarity between the third user vector and the three resource vectors.

[0187] In some embodiments, after inputting the second encoded vector into the second neural network and extracting user features and resource features from the second encoded vector in the second neural network to obtain the second user vector among multiple user vectors and the second resource vector among multiple resource vectors, the method further includes: determining a second similarity between the second user vector and the second resource vector; determining a second error value based on the actual interaction behavior of the user corresponding to the second user vector with the resource corresponding to the second resource vector and the second similarity; and adjusting the parameters of the second neural network based on the second error value.

[0188] In some embodiments, the recalled resources further include second recalled resources recalled based on the potential interests of the first user. The determining unit 1003 includes: a weight adjustment module 10031, used to increase the recommendation weight corresponding to the first recalled resource and the recommendation weight corresponding to the second recalled resource during the process of sorting and / or merging the recalled resources; and a determining module 10032, used to determine the resource to be recommended to the first user from the recalled resources according to the recommendation weight corresponding to the recalled resource.

[0189] In some embodiments, the apparatus for determining the resources to be recommended further includes: an interest point mining unit 1004, configured to determine the potential interest points of a first user; and a second recall unit 1005, configured to recall resources based on the potential interest points to obtain second recalled resources.

[0190] In some embodiments, the interest point mining unit 1004 includes at least one of the following: a first mining module 10041, configured to determine potential interest points as those where the first user's click count meets a requirement and the first user's most recent click time meets a time requirement; a second mining module 10042, configured to determine potential interest points among multiple popular interest points based on the similarity between the first user's user vector and the interest point vectors of multiple popular interest points, wherein the interest point vectors are determined based on the interest point interaction behavior data of multiple users; a third mining module 10043, configured to determine potential interest points as those related to the first user's search resources; and a fourth mining module 10044, configured to determine potential interest points as those from a third-party data platform.

[0191] Figure 9-10 The device for determining the resources to be recommended can execute the corresponding method embodiments described above. Its implementation principle and technical effect are similar, and will not be repeated here.

[0192] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0193] According to embodiments of this disclosure, this disclosure also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the scheme provided in any of the above embodiments.

[0194] According to embodiments of this disclosure, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the scheme provided in any of the above embodiments.

[0195] According to embodiments of this disclosure, this disclosure also provides a computer program product comprising: a computer program stored in a readable storage medium, at least one processor of an electronic device being able to read the computer program from the readable storage medium, and the at least one processor executing the computer program causing the electronic device to perform the scheme provided in any of the above embodiments.

[0196] Figure 11 This is a schematic block diagram of an example electronic device 1100 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0197] like Figure 11 As shown, the electronic device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) (e.g., ROM 1102) or a computer program loaded from a storage unit 1108 into a random access memory (RAM) (e.g., RAM 1103). The RAM 1103 may also store various programs and data required for the operation of the electronic device 1100. The computing unit 1101, ROM 1102, and RAM 1103 are interconnected via a bus 1104. An input / output (I / O) interface (e.g., I / O interface 1105) is also connected to the bus 1104.

[0198] Multiple components in electronic device 1100 are connected to I / O interface 1105, including: input unit 1106, such as keyboard, mouse, etc.; output unit 1107, such as various types of displays, speakers, etc.; storage unit 1108, such as disk, optical disk, etc.; and communication unit 1109, such as network card, modem, wireless transceiver, etc. Communication unit 1109 allows electronic device 1100 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0199] The computing unit 1101 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 performs the various methods and processes described above, such as the method for determining resources to be recommended. For example, in some embodiments, the method for determining resources to be recommended may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1108. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1100 via ROM 1102 and / or communication unit 1109. When the computer program is loaded into RAM 1103 and executed by the computing unit 1101, one or more steps of the method for determining resources to be recommended described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to perform a method for determining the resources to be recommended by any other suitable means (e.g., by means of firmware).

[0200] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard parts (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0201] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0202] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0203] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0204] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0205] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0206] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0207] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for determining a resource to be recommended, comprising: obtaining a user vector of a first user and a resource vector of a first resource from a plurality of user vectors and a plurality of resource vectors, wherein the plurality of user vectors and the plurality of resource vectors are determined according to a random walk corpus of a user-resource relationship graph; wherein the user-resource relationship graph comprises a plurality of nodes and edges; the plurality of nodes comprise user nodes and resource nodes; and the edges for connecting the user nodes and the resource nodes are constructed according to interaction behaviors between users and resources; the random walk corpus is a corpus for generating vectors obtained by performing weighted random walk on paths in the user-resource relationship graph; and the user-resource relationship graph is constructed by: obtaining resource interaction behavior data of a second user; determining an interaction frequency of the second user with a second resource according to the resource interaction behavior data of the second user; and determining a user node corresponding to the second user and a resource node corresponding to the second resource in the user-resource relationship graph, and determining a weight of an edge between the user node corresponding to the second user and the resource node corresponding to the second resource according to an inverse of the interaction frequency; determining whether to recall the first resource according to a similarity between the user vector of the first user and the resource vector of the first resource, to obtain a first recalled resource; determining a resource to be recommended for the first user according to the recalled resource, wherein the recalled resource comprises the first recalled resource.

2. The method of Claim 1, wherein, The plurality of user vectors and the plurality of resource vectors are obtained by: determining the user-resource relationship graph; performing weighted random walk in the user-resource relationship graph to obtain a random walk corpus; encoding the random walk corpus to obtain a first encoded vector; inputting the first encoded vector into a first neural network, and extracting user features and resource features of the first encoded vector in the first neural network to obtain a first user vector in the plurality of user vectors and a first resource vector in the plurality of resource vectors.

3. The method for determining a resource to be recommended according to claim 2, after the first encoded vector is input into the first neural network, and the user features and the resource features of the first encoded vector are extracted in the first neural network to obtain the first user vector in the plurality of user vectors and the first resource vector in the plurality of resource vectors, the method further comprises: determining a first similarity between the first user vector and the first resource vector; determining a first error value according to an actual interaction behavior of a user corresponding to the first user vector with a resource corresponding to the first resource vector and the first similarity; adjusting parameters of the first neural network according to the first error value.

4. The method for determining a resource to be recommended according to claim 1, wherein the plurality of user vectors and the plurality of resource vectors are obtained by: encoding resource interaction behavior data of a plurality of users to obtain a second encoded vector; inputting the second encoding vector into a second neural network, and performing user feature and resource feature extraction on the second encoding vector in the second neural network to obtain a second user vector in the plurality of user vectors and a second resource vector in the plurality of resource vectors; wherein a network parameter of the second neural network is related to a similarity between a third user vector generated by a third neural network and a third resource vector generated by the third neural network, and the greater the similarity between the third user vector and the third resource vector, the smaller the network parameter of the second neural network.

5. The method of claim 4, before the inputting the second encoding vector into the second neural network, further comprising: inputting the second encoding vector into the third neural network, and performing user feature and resource feature extraction on the second encoding vector in the third neural network to obtain the third user vector and the third resource vector; updating the network parameter of the second neural network according to an inverse of the similarity between the third user vector and the third resource vector.

6. The method of claim 5, wherein, The updating the network parameter of the second neural network according to the inverse of the similarity between the third user vector and the third resource vector comprises: updating a network parameter of a random sampling layer in the second neural network to the inverse of the similarity between the third user vector and the third resource vector.

7. The method of claim 4, after the inputting the second encoding vector into the second neural network, and performing user feature and resource feature extraction on the second encoding vector in the second neural network to obtain a second user vector in the plurality of user vectors and a second resource vector in the plurality of resource vectors, further comprising: determining a second similarity between the second user vector and the second resource vector; determining a second error value according to an actual interaction behavior of a user corresponding to the second user vector on a resource corresponding to the second resource vector and the second similarity; performing parameter adjustment on the second neural network according to the second error value. 8.The method for determining resources to be recommended according to any one of claims 1-7, wherein, The recall resources further include second recall resources recalled according to potential interest points of the first user, and the determining the to-be-recommended resources for the first user according to the recall resources comprises: increasing a recommendation weight corresponding to the first recall resource and a recommendation weight corresponding to the second recall resource in a process of resource sorting and / or resource fusion on the recall resources; determining the to-be-recommended resources for the first user from the recall resources according to the recommendation weight corresponding to the recall resources.

9. The method of claim 8, further comprising: determining potential interest points of the first user; performing resource recall according to the potential interest points to obtain the second recall resources.

10. The method of Claim 9, wherein, The determining the potential interest points of the first user comprises at least one of the following: determining the potential interest points as interest points whose click times of the first user satisfy a time requirement and whose latest click times of the first user satisfy a time requirement. determine the potential interest point from the plurality of popular interest points according to a similarity between the user vector of the first user and interest point vectors of the plurality of popular interest points, wherein the interest point vectors are determined according to interest point interaction behavior data of a plurality of users; determine the potential interest point as an interest point related to a search resource of the first user; determine the potential interest point as an interest point from a third-party data platform.

11. A determination apparatus of a resource to be recommended, comprising: an acquisition unit, configured to acquire, from a plurality of user vectors and a plurality of resource vectors, a user vector of a first user and a resource vector of a first resource, wherein the plurality of user vectors and the plurality of resource vectors are determined according to a random walk corpus of a user-resource relationship graph; a first recall unit, configured to determine whether to recall the first resource according to a similarity between the user vector of the first user and the resource vector of the first resource, to obtain a first recalled resource; a determination unit, configured to determine, according to the recalled resource, a resource to be recommended for the first user, wherein the recalled resource includes the first recalled resource; wherein the user-resource relationship graph includes a plurality of nodes and edges; the plurality of nodes include user nodes and resource nodes; and an edge for connecting a user node and a resource node is constructed according to an interaction behavior between a user and a resource; the random walk corpus is a corpus for generating vectors obtained by performing weighted random walk on a path in the user-resource relationship graph; and the user-resource relationship graph is constructed by: acquiring resource interaction behavior data of a second user; determining, according to the resource interaction behavior data of the second user, an interaction frequency of the second user with a second resource; determining, in the user-resource relationship graph, a user node corresponding to the second user and a resource node corresponding to the second resource, and determining a weight of an edge between the user node corresponding to the second user and the resource node corresponding to the second resource according to an inverse of the interaction frequency. 12.The apparatus for determining a resource to be recommended according to claim 11, wherein, The plurality of user vectors and the plurality of resource vectors are obtained by: determining the user-resource relationship graph; performing weighted random walk in the user-resource relationship graph to obtain a random walk corpus; encoding the random walk corpus to obtain a first encoded vector; inputting the first encoded vector into a first neural network, and extracting user features and resource features of the first encoded vector in the first neural network to obtain a first user vector in the plurality of user vectors and a first resource vector in the plurality of resource vectors.

13. The determination apparatus of a resource to be recommended according to claim 12, after the first encoded vector is input into the first neural network, and user features and resource features of the first encoded vector are extracted in the first neural network to obtain a first user vector in the plurality of user vectors and a first resource vector in the plurality of resource vectors, the determination apparatus further comprises: determining a first similarity between the first user vector and the first resource vector. determine a first error value according to actual interaction behavior of a user corresponding to the first user vector to a resource corresponding to the first resource vector and the first similarity; adjust parameters of the first neural network according to the first error value.

14. The device of claim 11, wherein the plurality of user vectors and the plurality of resource vectors are obtained by: encoding resource interaction behavior data of a plurality of users to obtain a second encoding vector; inputting the second encoding vector into a second neural network, and extracting user features and resource features of the second encoding vector in the second neural network to obtain a second user vector in the plurality of user vectors and a second resource vector in the plurality of resource vectors; wherein, network parameters of the second neural network are related to a similarity between a third user vector generated by a third neural network and a third resource vector generated by the third neural network, and the greater the similarity between the third user vector and the third resource vector, the smaller the network parameters of the second neural network.

15. The device of claim 14, wherein before the second encoding vector is input into the second neural network, the method further comprises: inputting the second encoding vector into the third neural network, and extracting user features and resource features of the second encoding vector in the third neural network to obtain the third user vector and the third resource vector; updating the network parameters of the second neural network according to an inverse of the similarity between the third user vector and the third resource vector.

16. The apparatus according to claim 15, wherein The updating the network parameters of the second neural network according to the inverse of the similarity between the third user vector and the third resource vector comprises: updating network parameters of a random sampling layer in the second neural network to the inverse of the similarity between the third user vector and the third resource vector.

17. The device of claim 14, wherein after the second encoding vector is input into the second neural network, and extracting user features and resource features of the second encoding vector in the second neural network to obtain a second user vector in the plurality of user vectors and a second resource vector in the plurality of resource vectors, the method further comprises: determining a second similarity between the second user vector and the second resource vector; determining a second error value according to actual interaction behavior of a user corresponding to the second user vector to a resource corresponding to the second resource vector and the second similarity; adjusting parameters of the second neural network according to the second error value. 18.The apparatus for determining resources to be recommended according to any one of claims 11-17, wherein, The recalled resources further comprise second recalled resources recalled according to potential interest points of the first user, and the determining unit comprises: a weight adjusting module configured to increase a recommendation weight corresponding to the first recalled resources and a recommendation weight corresponding to the second recalled resources in a process of resource sorting and / or resource fusion of the recalled resources. determining, according to the recommendation weight corresponding to the recall resource, a to-be-recommended resource for the first user from the recall resource. 19.The to-be-recommended resource determining apparatus of claim 18, further comprising: a point of interest mining unit configured to determine a potential point of interest of the first user; a second recall unit configured to perform resource recall according to the potential point of interest to obtain the second recall resource.

20. The apparatus according to claim 19, wherein The point of interest mining unit comprises at least one of: a first mining module configured to determine that the potential point of interest is a point of interest whose click number of the first user meets a number requirement and whose latest click time meets a time requirement; a second mining module configured to determine the potential point of interest from a plurality of popular points of interest according to a similarity between a user vector of the first user and a point of interest vector of the popular points of interest, wherein the point of interest vector is determined according to point of interest interaction behavior data of a plurality of users; a third mining module configured to determine that the potential point of interest is a point of interest related to a search resource of the first user; a fourth mining module configured to determine that the potential point of interest is a point of interest from a third-party data platform. 21.An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of determining a to-be-recommended resource according to any one of claims 1-10.

22. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method of determining a to-be-recommended resource according to any one of claims 1-10. 23.A computer program product comprising a computer program which, when executed by a processor, implements the steps of the method of determining a to-be-recommended resource according to any one of claims 1-10.

Images