Resource Recommendation Method Based on Neighborhood Aggregation and Related Devices

Abstract

This application relates to the field of artificial intelligence, and discloses a resource recommendation method and related devices based on neighborhood aggregation. The method includes: performing social aggregation on the neighborhood influence vector of a target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user; generating the target feature vector of the target user according to the resource embedding vector of the historical preference resources corresponding to the target user and the target embedding vector of the target user; the resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource; determining the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set; screening resources in the resource set according to the preference weights to determine the target resources recommended to the target user. This solution improves the accuracy of determining the target resources for the target user.

Description

[0001] Technical Field

[0002] This application relates to the field of artificial intelligence technology, and more specifically, to a resource recommendation method based on neighborhood aggregation and related devices. Background Art

[0003] With the rise and rapid popularization of online social networks, users on social platforms can share their opinions and life experiences on social platforms such as Weibo, Facebook, WeChat, etc., and build their own social circles. This gives rise to the problem of social recommendation. In related technologies, the resources recommended to users based on social recommendation have the problem of low accuracy, resulting in users needing to perform secondary resource screening or make secondary resource requests, thus causing low resource acquisition efficiency for users. Summary of the Invention

[0004] In view of the above problems, embodiments of the present application propose a resource recommendation method based on neighborhood aggregation and related devices to improve the accuracy of resource recommendation.

[0005] According to one aspect of the embodiments of the present application, a resource recommendation method based on neighborhood aggregation is provided, including: performing social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user; the initial embedding vector is determined according to the user attribute information of the corresponding user and the preferred resource information of the corresponding user; generating the target feature vector of the target user according to the resource embedding vector of the historical preferred resources corresponding to the target user and the target embedding vector of the target user; the resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource; determining the preference weight of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set; screening resources in the resource set according to the preference weight to determine the target resources recommended to the target user.

[0006] According to one aspect of the embodiments of the present application, a resource recommendation device based on neighborhood aggregation is provided, including: an aggregation module, configured to perform social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user; the initial embedding vector is determined according to the user attribute information of the corresponding user and the preferred resource information of the corresponding user; a target feature vector generation module, configured to generate a target feature vector of the target user according to the resource embedding vector of the historical preferred resources corresponding to the target user and the target embedding vector of the target user; the resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource; a preference weight determination module, configured to determine the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of the resources in the resource set; a target resource determination module, configured to perform resource screening in the resource set according to the preference weights to determine the target resources recommended to the target user.

[0007] According to one aspect of the embodiments of the present application, an electronic device is provided, including: a processor; a memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, the resource recommendation method based on neighborhood aggregation as described above is implemented.

[0008] According to one aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the resource recommendation method based on neighborhood aggregation as described above is implemented.

[0009] In this solution, the target feature vector of the target user is generated by combining the features in four dimensions, namely, the user attribute information of the target user, the corresponding preferred resource information, the set of neighborhood users of the target user, and the historical preferred resources of the target user. It can reflect the features of the target user in multiple dimensions, and the features expressed by the target user are more comprehensive and accurate. Based on this, determining the target resources recommended to the target user according to the target feature vector of the target user and the resource embedding vectors of each resource can ensure a higher matching degree between the obtained target resources and the target user, and ensure the accuracy of resource recommendation to the target user. It can greatly reduce the probability that the target user performs resource screening again among the recommended resources due to low resource recommendation accuracy or makes a secondary resource request due to inaccurate resource recommendation. Therefore, the resource acquisition efficiency of the target user is effectively improved.

[0010] Moreover, by performing social aggregation based on the initial embedding vector of the target user and the neighborhood influence vectors of the set of neighboring users corresponding to the target user, the influence of the social network where the target user is located on the target user's attitude towards resources is fully explored, improving the performance of social recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings herein are incorporated into and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0012] Figure 1 FIG. shows a schematic diagram of an exemplary system architecture to which the technical solution of the embodiment of the present application can be applied.

[0013] Figure 2 is a flowchart of a resource recommendation method based on neighborhood aggregation according to an embodiment of the present application.

[0014] Figure 3 is a flowchart of step 210 according to an embodiment of the present application.

[0015] Figure 4 is a flowchart of step 320 according to an embodiment of the present application.

[0016] Figure 5 is a flowchart of the steps before step 210 according to an embodiment of the present application.

[0017] Figure 6 is a flowchart of the steps before step 230 according to an embodiment of the present application.

[0018] Figure 7 is a flowchart of the steps before step 210 according to another embodiment of the present application.

[0019] Figure 8 is a flowchart of step 230 according to an embodiment of the present application.

[0020] Figure 9 is a flowchart of a resource recommendation method based on neighborhood aggregation according to a specific embodiment of the present application.

[0021] Figure 10 is a block diagram of a resource recommendation device based on neighborhood aggregation according to an embodiment of the present application.

[0022] Figure 11 FIG. shows a schematic diagram of the structure of a computer system of an electronic device suitable for implementing the embodiment of the present application. Detailed Implementation Modes

[0023] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0024] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of this application. However, those skilled in the art will realize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be used. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this application.

[0025] The block diagrams shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0026] The flowcharts shown in the drawings are merely illustrative and do not necessarily include all the content and operations / steps, nor do they necessarily have to be executed in the order described. For example, some operations / steps can be decomposed, while some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0027] It should be noted that: "a plurality" as mentioned herein refers to two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after.

[0028] Figure 1 A schematic diagram of an exemplary system architecture to which the technical solutions of the embodiments of this application can be applied is shown.

[0029] As Figure 1 shown, the system architecture may include terminal devices (such as Figure 1One or more of the smart phone 101, tablet computer 102, and portable computer 103 shown (which can of course also be a desktop computer, etc.), network 104, and server 105. The network 104 is used to provide a medium for a communication link between the terminal device and the server 105. The network 104 can include various connection types, such as wired communication links, wireless communication links, and so on.

[0030] The terminal device can run a social application client for interacting with friends, forwarding and sharing information, posting information, etc. based on the social application client. The server 105 is used to provide services for the social application client.

[0031] In this application, the server 105 can further collect information about user information and resources, where the resources can be articles, blogs, news, videos, advertisements, products, application programs, service programs, official accounts, merchants, etc. The user information collected can include the basic attribute information of the user and the social information of the user, and the information of the resources collected can include the basic attribute information of the resources and the interaction information of the user with the resources. On this basis, the server 105 can determine resource recommendations for the user according to the solution of this application based on the user information and the resource information.

[0032] It should be understood that Figure 1 the numbers of the terminal devices, networks, and servers in are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. For example, the server 105 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, and big data and artificial intelligence platforms.

[0033] The implementation details of the technical solution of the embodiments of this application are elaborated in detail below:

[0034] Figure 2 shows a flowchart of a resource recommendation method based on neighborhood aggregation according to an embodiment of this application. This method can be executed by a computer device with processing capabilities, such as a server, etc., and is not specifically limited here. Referring to Figure 2 as shown, this method at least includes steps 210 to 240, which are introduced in detail as follows:

[0035] Step 210, perform social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user; the initial embedding vector is determined according to the user attribute information of the corresponding user and the preferred resource information of the corresponding user.

[0036] The target user generally refers to a user for whom resource recommendation is to be performed. The target user can be a user in a social application or other applications integrating social functions, etc., and no specific limitation is made here.

[0037] The user attribute information may include basic attributes such as the user's age, location, gender, occupation, etc. The user attribute information can be obtained from the user's registration information in the application.

[0038] The user's preferred resource information is used to indicate the resources preferred by the user. In some embodiments, the user's preferred resources can be determined by the interaction behavior between the user and the published resources. For example, the user's preferred resources can be the resources that the user has given good reviews, liked, collected, shared, purchased, etc. In some embodiments, the user's preferred resources can also be the resources that match the user's interest tags. Among them, the user's interest tags can be tags added to the user based on the content published by the user, the content shared by the user, etc.

[0039] In some embodiments of the present application, the user attribute information of the target user can be vectorized to obtain the first feature vector of the target user; and the user's preferred resource information is vectorized to obtain the second feature vector of the target user; then, the first feature vector and the second feature vector of the target user are fused to obtain the initial embedding vector of the target user.

[0040] In some embodiments of the present application, the first feature vector and the second feature vector can be fused through a multi-layer perceptron network. The multi-layer perceptron network introduces one or more hidden layers on the basis of a single-layer neural network, and the hidden layer is located between the input layer and the output layer. The multi-layer perceptron network is a neural network composed of fully connected layers with at least one hidden layer, and the output of each hidden layer is transformed through an activation function. The number of layers of the multi-layer perceptron and the number of hidden units in each hidden layer are both hyperparameters.

[0041] The set of neighborhood users of the target user is a set formed by the neighborhood users of the target user. In some embodiments, the neighborhood users of the target user can be users with a relatively high degree of association with the target user. Users with a relatively high degree of association with the target user, such as the target user's friends, users with a relatively high interaction frequency with the target user, users of the accounts followed by the target user (such as the public account, blogger, etc. followed), users whose preferred resource types are similar to or the same as those of the target user, users whose preferred resources have a similarity higher than the set similarity threshold with those of the target user, etc., and no specific limitation is made here.

[0042] In some other embodiments of the present application, the neighboring users of the target user may also be users who are associated with the target user and whose trust score with the target user is not lower than a set trust score threshold.

[0043] In some embodiments of the present application, the association degree between the target user and other users can be calculated from multiple dimensions. For example, the association degree between two users can be calculated from the friend dimension and the preferred resource dimension. Another example is to calculate the association degree between two users from the friend dimension, the preferred resource dimension, and the interaction frequency dimension. In some embodiments, the neighboring users of the target user may also be users who have a friendship relationship with the target user and whose types of preferred resources are similar to or have a relatively high similarity with those of the target user.

[0044] The neighborhood influence vector of the set of neighboring users corresponding to the target user represents the characteristics of all neighboring users in the set of neighboring users. In some embodiments of the present application, the neighborhood influence vector can be obtained by aggregating the embedding vectors of all neighboring users in the set of neighboring users corresponding to the target user.

[0045] The sets of neighboring users of different users are different. Correspondingly, the corresponding neighborhood influence vectors are also different. For a user, the social circle where the user is located will affect the user's attitude towards resources, and the influence of the social circle on the user can be reflected by the characteristics of the users associated with the user in the social circle where the user is located. Therefore, social aggregation of the neighborhood influence vector of the set of neighboring users corresponding to the target user and the initial embedding vector of the target user realizes the aggregation of the characteristics of the target user and the set of neighboring users corresponding to the target user, so that the obtained target embedding vector of the target user can not only reflect the characteristics of the target user but also reflect the characteristics of the set of neighboring users corresponding to the target user. Therefore, the target embedding vector introduces the influence of the set of neighboring users on the target user.

[0046] Step 220, generate a target feature vector of the target user according to the resource embedding vector of the historical preferred resource corresponding to the target user and the target embedding vector of the target user; the resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource.

[0047] The resource attribute information can be the basic attributes of the resource. For example, the name of the resource, the release time of the resource, the publisher of the resource, the category to which the resource belongs, the theme of the resource, the abstract of the resource, etc.; for text-type resources, the basic attributes of the resource can also include the number of words in the resource; for audio or video-type resources, the basic attributes of the resource can also include the duration of the audio or video. It is worth mentioning that for different types of resources, the classification principles of the resources may vary. For example, if the resource is news, the classification principle of the resource can be the theme of the resource, such as entertainment, education, technology, etc.

[0048] The associated user information of the resource can be used to indicate the target users of the resource, or to indicate the users whose favorable reviews of the resource are higher than the first set favorable review threshold. Among them, the target users of the resource can be determined according to the target user conditions set by the resource publisher. The favorable review of the resource can be calculated based on the user feedback behavior of the resource during the period after it is released. The user feedback behavior of the resource includes, for example, user behaviors such as forwarding the resource, liking the resource, rating, giving a favorable review, sharing, and purchasing.

[0049] In some embodiments of the present application, the resource attribute information of the resource can be vectorized to obtain the first resource feature vector of the resource; and the associated user information of the resource attribute information can be vectorized to obtain the second resource feature vector of the associated user information; then, the first resource feature vector and the second resource feature vector of the associated user information are fused to obtain the resource embedding vector of the resource.

[0050] The historical preference resources corresponding to the target user can be the resources for which the target user has given a favorable review higher than the second set favorable review threshold. Similarly, the favorable review given by the target user to the resource can be calculated based on the user feedback behavior of the target user on the resource during the period after the resource is released. For example, if the resource is a commodity, the favorable review of the target user for the commodity can be calculated based on the user behaviors such as the comments, ratings, and whether to forward the commodity link made by the target user after purchasing the commodity. If the resource is a public account article, the favorable review of the target user for the commodity can be calculated based on the user behaviors such as the message comments, likes, forwards, and shares of the target user after reading the public account article.

[0051] In some embodiments of the present application, the resource embedding vector of the historical preference resources corresponding to the target user and the target embedding vector of the target user can be fused to obtain the target feature vector of the target user. The fusion performed can be to vectorially superimpose the resource embedding vector of the historical preference resources corresponding to the target user and the embedding vector of the target user, and use the superimposed result as the target feature vector of the target user.

[0052] Since the historical preference resources corresponding to the target user can reflect the characteristics of the target user, such as the preference characteristics for resources, therefore, fusing the resource embedding vector of the historical preference resources corresponding to the target user and the target embedding vector of the target user can ensure that the obtained target feature vector of the target user can further reflect the characteristics of the target user in the dimension of historical preference resources.

[0053] Step 230, according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set, determine the preference weights of the target user for each resource in the resource set.

[0054] The preference weight is used to indicate the degree of interest of the target user in the resource. The higher the preference weight, the higher the degree of interest of the target user in the resource, and thus the higher the matching degree between the resource and the target user.

[0055] In some embodiments of the present application, the preference weights of the target user for each resource can be automatically predicted through a trained neural network model (such as the preference prediction model in the following text).

[0056] Step 240, according to the preference weights, perform resource screening in the resource set to determine the target resources recommended to the target user.

[0057] In some embodiments of the present application, the resources in the resource set can be sorted according to the order of preference weights from high to low, and then the resources ranked in the top set number in the sorting are determined as the target resources recommended to the target user. In another embodiment, based on a set preference weight threshold, the resources in the resource set with a preference weight higher than the preference weight threshold relative to the target user can be determined as the target resources recommended to the target user.

[0058] After determining the target resources recommended to the target user, when a resource recommendation request initiated by the target user is received, push the target resources to the user.

[0059] In this solution, the target feature vector of the target user combines features in four dimensions: the user attribute information of the target user, the corresponding preferred resource information, the set of neighboring users of the target user, and the historical preferred resources of the target user, to generate the target feature vector of the target user. As a result, the target feature vector of the target user can reflect the characteristics of the target user in multiple dimensions, and the characteristics expressed by the target user are more comprehensive and accurate. Based on this, the target resources recommended to the target user are determined according to the target feature vector of the target user and the resource embedding vectors of each resource, which can ensure a higher matching degree between the obtained target resources and the target user, and ensure the accuracy of resource recommendation to the target user. It can greatly reduce the probability that the target user will screen resources again among the recommended resources or make a secondary resource request due to inaccurate resource recommendation. Therefore, the resource acquisition efficiency of the target user is effectively improved.

[0060] In some embodiments of the present application, as Figure 3 shown, step 210 includes: step 310, obtaining the k-level embedding vector of the target user obtained by k-level social aggregation; wherein, the initial embedding vector of the target user is used for first-level social aggregation; k is a positive integer. And step 320, obtaining the (t + 1)-level neighborhood influence vector for the (t + 1)-level social aggregation.

[0061] Step 330, performing (k + 1)-level social aggregation on the k-level embedding vector of the target user and the (k + 1)-level neighborhood influence vector to obtain the (k + 1)-level embedding vector of the target user.

[0062] In the solution of this embodiment, for each user, the initial embedding vector of the user and the neighborhood influence vector of the set of neighboring users corresponding to the user on the target user are subjected to social aggregation level by level to obtain the embedding vectors of each level of the user. On this basis, for the target user, the k-level embedding vectors of each neighboring user in the neighborhood set of the target user can be aggregated correspondingly to obtain the (k + 1)-level neighborhood influence vector for the (k + 1)-level social aggregation.

[0063] In some embodiments of the present application, the (k + 1)-level social aggregation performed may be to vector-connect the k-level embedding vector of the target user and the corresponding (k + 1)-level neighborhood influence vector, that is, to splice the two vectors into one vector, and use the vector obtained by the vector connection as the (k + 1)-level embedding vector of the target user.

[0064] Step 340, determining whether the social aggregation level (k + 1) reaches the set level threshold.

[0065] If the social aggregation level (k + 1) reaches the set level threshold, then step 350 is executed: Use the (k + 1)-level embedding vector of the target user as the target embedding vector of the target user.

[0066] If the social aggregation level (k + 1) is less than the set level threshold, then step 360 is executed: Continue to perform the next-level social aggregation on the (k + 1)-level embedding vector of the target user.

[0067] The set level threshold can be set as needed and is not specifically limited here. If the current level of social aggregation (i.e., the social aggregation level) reaches the set level threshold, then the social aggregation ends, and the (k + 1)-level embedding vector of the target user obtained from the current social aggregation is used as the target embedding vector of the target user; otherwise, continue to use the (k + 1)-level embedding vector of the target user for the next-level social aggregation.

[0068] By performing social aggregation level by level, the global social influence of the set of neighborhood users of the target user on the target user can be accurately mined. Moreover, since the neighborhood influence vector is generated based on the embedding vectors of each neighborhood user in the set of neighborhood users, the embedding vectors of neighborhood users are generated based on the user attribute information and corresponding preference resource information of neighborhood users, and the initial embedding vector of the target user is also generated based on the user attribute information and preference resource information of the target user. Therefore, performing social aggregation level by level is equivalent to realizing the deep interaction between users and resources, can effectively mine the features reflecting the user's attitude towards resources, and can ensure the accuracy of subsequent resource recommendations.

[0069] In some embodiments of the present application, as Figure 4 shown, step 320 includes: step 410, obtaining the t-level embedding vectors of each neighborhood user in the neighborhood set. Step 420, aggregating the attention coefficients of all neighborhood users in the neighborhood set with the t-level embedding vectors of all neighborhood users in the neighborhood set to obtain the (t + 1)-level neighborhood influence vector corresponding to the (t + 1)-level social aggregation; among them, the 1-level neighborhood influence vector corresponding to the 1-level social aggregation is calculated based on the initial embedding vectors of each neighborhood user in the neighborhood set.

[0070] The attention coefficient is used to characterize the influence degree of neighborhood users. In some embodiments of the present application, the attention scores of each neighborhood user can be calculated first based on the hyperbolic tangent function and the trained weight matrix, and then the attention scores of each neighborhood user are normalized to obtain the attention scores of each user. On this basis, aggregation is performed based on the attention scores of each neighborhood user and the embedding vectors of each neighborhood user at each level, so that the corresponding neighborhood influence vectors at each level can reflect the heterogeneous connections of neighborhood users in the set of neighborhood users corresponding to the target user to the target user.

[0071] In some embodiments of the present application, such as Figure 5 shown, before step 210, the method further includes: step 510, generating a first feature vector of the target user according to the user attribute information of the target user; and step 520, generating a second feature vector of the target user according to the preference resource information of the target user.

[0072] In some embodiments of the present application, a pre-trained first embedding layer is used to map the preference resource information of the target user into a low-dimensional first one-hot vector representation (this process is to perform one-hot encoding on the preference resource information of the target user), and then the first one-hot vector representation is transformed through a trained parameter matrix to obtain the second feature vector of the target user.

[0073] One-hot encoding, also known as one-hot coding or one-hot representation, is a method of using an N-bit status register to encode N states. Each state has its own independent register bit, and at any time, only one of them is valid. In other words, for each feature, if it has m possible values, then after one-hot encoding, it becomes m binary features, and these features are mutually exclusive, and only one is activated at a time. For example, the feature of grades has 3 state values: good, medium, and poor, and after one-hot encoding, it becomes 100, 010, and 001.

[0074] Similarly, the same method can be used to generate the first feature vector of the target user based on the user attribute information of the target user, which will not be elaborated here.

[0075] Step 530, fusing the first feature vector of the target user and the second feature vector of the target user to obtain the initial embedding vector of the target user.

[0076] In some embodiments of the present application, the first feature vector and the second feature vector of the target user can be fused through a trained multi-layer perceptron network. Specifically, step 530 includes: vector-connecting the first feature vector of the target user and the second feature vector of the target user to obtain a user connection vector; inputting the user connection vector into a first multi-layer perceptron network, and the first multi-layer perceptron network outputs the initial embedding vector of the target user according to the user connection vector.

[0077] Through the multi-layer perceptron network, effective fusion of the first resource feature vector and the second resource feature vector of the resource is achieved.

[0078] In some embodiments of the present application, such as Figure 6As shown, before step 230, the method further includes: step 610, generating a first resource feature vector corresponding to each resource according to the resource attribute information of each resource in the resource set; and step 620, generating a second resource feature vector corresponding to each resource according to the associated user information of each resource in the resource set.

[0079] In some embodiments of the present application, a pre-trained second embedding layer is used to map the preference resource information of the target user into a low-dimensional second one-hot vector representation, and then the second one-hot vector representation is transformed through a second parameter matrix to obtain the second resource feature vector of the resource.

[0080] Similarly, the first resource feature vector of the resource can be generated based on the same method according to the resource attribute information of the resource, which will not be elaborated here.

[0081] Step 630, fusing the first resource feature vector corresponding to the resource and the second resource feature vector corresponding to the resource to obtain the resource embedding vector corresponding to the resource.

[0082] Similar to the initial embedding vector of the target user, the first resource feature vector and the second resource feature vector can be fused through a trained multi-layer perceptron network. Specifically, step 630 includes: vector connecting the first resource feature vector corresponding to the resource and the second resource feature vector corresponding to the resource to obtain a resource connection vector; inputting the resource connection vector into a second multi-layer perceptron network, and the second multi-layer perceptron network outputs the initial embedding vector of the resource according to the resource connection vector.

[0083] Through the multi-layer perceptron network, effective fusion of the first resource feature vector and the second resource feature vector of the resource is achieved.

[0084] In some embodiments of the present application, as Figure 7 shown, before step 210, the method further includes:

[0085] Step 710, calculating the social trust degree between the target user and each candidate user in the candidate user set according to the social interaction information of the target user.

[0086] The social trust degree between the target user and the candidate user is a quantitative representation of the trust degree of the target user in the candidate user. The social interaction information of the target user is used to indicate the interaction behavior of the target user with other users, and the social interaction behavior is, for example, the chat behavior based on the chat window, the behavior of commenting, liking on the published content (or the forwarded content), the behavior of replying to the comments of the other party, etc. Similarly, the social interaction information of the candidate user is used to indicate the social interaction behavior of the candidate user with other users.

[0087] Based on the social interaction information of the target user, the number of various interaction behaviors (such as the interaction behavior of commenting on the published content as mentioned above, the chatting behavior based on the chat window, etc.) between the target user and the candidate user can be determined. Then, the number of various interaction behaviors between the target user and the candidate user can be weighted to obtain the total interaction times of the interaction behaviors between the target user and the candidate user. At the same time, according to the same method, based on the social interaction information of the target user, the total interaction times between the target user and other users can also be statistically determined.

[0088] In some embodiments of the present application, the social trust degree between the target user and any candidate user can also be calculated only according to the social interaction information of the target user. The social trust degree between the target user and a candidate user can be calculated according to the following formula:

[0089]

[0090] where, H i,j represents the social trust degree of the target user i for the candidate user j; D i,j represents the total interaction times between the target user i and the candidate user j; represents the minimum interaction times between the target user i and other users; represents the maximum interaction times between the target user i and other users.

[0091] In some other embodiments of the present application, the interaction frequency between the target user and each user can be further calculated based on the interaction times and interaction time between the target user and each user. Based on the interaction frequency between the target user and each user, the social trust degree between the target user and each candidate user is calculated. For example, the social trust degree between the target user and the candidate user can be calculated according to the following formula:

[0092]

[0093] where, f i,j represents the interaction frequency between the target user i and the candidate user j; represents the median of the interaction frequencies between the target user i and each user.

[0094] In some embodiments of the present application, the candidate users in the candidate user set can be the friend users of the target user, or the friend users and followed users of the target user, etc., and no specific limitation is made here.

[0095] Step 720, calculate the resource preference similarity between the target user and each candidate user in the candidate user set according to the preference resource information of the target user and the preference resource information of each candidate user in the candidate user set.

[0096] The resource preference similarity between the target user and the candidate user is used to indicate the similarity between the target user and the candidate user in terms of preferred resources.

[0097] Based on the preferred resource information of the target user and the preferred resource information of the candidate user, the common preferred resources of the target user and the candidate user can be determined. In one embodiment, based on the number of common preferred resources of the target user and the candidate user and the total number of the target user's preferred resources, the resource preference similarity between the target user and the candidate user is calculated. In another embodiment, based on the similarity weights of each preferred resource, the similarity weights corresponding to the common preferred resources of the target user and the candidate user can be weighted, and the weighted result is used as the resource preference similarity between the target user and the candidate user.

[0098] Step 730, according to the social trust degree and resource preference similarity between the target user and each candidate user in the candidate user set, screen the candidate users in the candidate user set.

[0099] Step 740, add the screened candidate users to the set of neighborhood users corresponding to the target user.

[0100] In some embodiments of the present application, for each candidate user, the social trust degree and resource preference similarity between the target user and the candidate user can be weighted, and the weighted result is used as the comprehensive score between the target user and the candidate user. Then, based on this comprehensive score, screen the candidate users in the candidate user set.

[0101] In some embodiments of the present application, candidate users with a comprehensive score higher than the set score threshold can be added to the set of neighborhood users corresponding to the target user. Further, the number of neighborhood users (specified number) in the set of neighborhood users can also be set. Thus, the candidate users are sorted according to the comprehensive score from high to low, and the candidate users ranked in the top specified number are added to the set of neighborhood users of the target user.

[0102] Through the above process, the set of neighborhood users corresponding to the target user is determined based on the social trust degree and preferred resource similarity between the target user and the candidate user, so as to ensure the social association degree and resource association degree between the target user and the neighborhood users. The higher the association degree between a user and the target user, the greater the influence of this user on the target user, ensuring that the set of neighborhood users screened and determined based on the social trust degree and preferred resource similarity with the target user can accurately reflect the characteristics of the social circle where the target user is located.

[0103] In some embodiments of the present application, such as Figure 8As shown in the figure, step 230 includes: step 810, vector-connecting the target feature vector of the target user with the resource embedding vectors of each resource in the resource set to obtain the connection feature vector of the target user with respect to each resource. Step 820, inputting the connection feature vector of the target user with respect to each resource into the preference prediction model. Step 830, the preference prediction model predicts the preference weights according to the connection feature vector of the target user with respect to each resource, and outputs the preference weights of the target user for each resource.

[0104] The preference prediction model is a model constructed by a neural network. The neural network can be, for example, a fully connected network, a recurrent neural network, a long short-term memory neural network, etc., which are not specifically limited here.

[0105] To ensure the prediction accuracy of the preference prediction model, before step 820, the preference prediction model also needs to be trained with training data. The training data can include the target feature vectors of sample users, the resource embedding vectors of sample resources, and the preference scores of sample users for each sample resource; then, input the target feature vector of a sample user and the resource embedding vector of a sample resource into the preference prediction model, and the preference prediction model predicts the preference weights based on the target feature vector of the sample user and the resource embedding vector of the sample resource, and outputs the predicted preference weight of the sample user for the sample resource; if the deviation between the predicted preference weight and the preference score of the sample user for the sample resource is not within the preset range, adjust the parameters of the preference prediction model, and re-predict the predicted preference weight between the sample user and the sample resource through the adjusted preference prediction model until the deviation between the predicted preference weight and the preference score of the sample user for the sample resource is within the preset range. By training the preference prediction model, the accuracy of the preference prediction model for predicting preference weights can be ensured.

[0106] Figure 9 It is a flowchart of a resource recommendation method based on neighborhood aggregation shown in a specific embodiment of the present application. As Figure 9 shown, if user a is used as the target user, fuse the first feature vector and the second feature vector of user a to obtain the initial embedding vector of user a.

[0107] Specifically, the fusion of the first feature vector and the second feature vector of user a can be carried out according to the following process:

[0108]

[0109] Among them, represents the initial embedding vector of user a; x a represents the first feature vector of user a; p aThe second feature vector representing user a; g u Represents the first multi-layer perceptron network; Represents vector concatenation of two vectors.

[0110] Similarly, for resources (resources to be recommended, historical preference resources of user a, etc.), the resource embedding vector of the resource can be determined according to the following process:

[0111]

[0112] Among them, v i Represents the resource embedding vector of resource i; y i Represents the first resource feature vector of resource i; q i Represents the second resource feature vector of resource i; g v Represents the second multi-layer perceptron network for fusing the first resource feature vector and the second resource feature vector.

[0113] Then, social aggregation is performed step by step based on the initial embedding vector of user a. Before that, it is necessary to first calculate the influence vectors of each level of the neighborhood users corresponding to user a on user a. Specifically, first calculate the attention scores of each neighborhood user in the set of neighborhood users corresponding to user a. The attention scores of each neighborhood user can be calculated according to the following formula:

[0114]

[0115] Among them, T a Represents the set of neighborhood users corresponding to user a; b represents a neighborhood user in the set of neighborhood users corresponding to user a; W b Represents the weight matrix; Represents the initial embedding vector of neighborhood user b; α b Represents the attention score of neighborhood user b.

[0116] Then, the attention scores of the neighborhood users are normalized. In a specific embodiment, the normalization can be performed through the softmax function. Specifically, the normalization can be performed according to the following formula to obtain the attention coefficients of each neighborhood user:

[0117]

[0118] Among them, ρ b Represents the attention coefficient of neighborhood user b.

[0119] Then, the influence vectors of each level of the neighborhood are calculated based on the attention coefficients of each neighborhood user and the embedding vectors of each level of the neighborhood users. Specifically, the influence vector of the (t + 1)-th level of the neighborhood can be calculated according to the following formula:

[0120]

[0121] Among them, represents the set of neighboring users \(T\) of user \(a\); a is the \((t + 1)\)-th level neighborhood influence vector; \(\sigma\) represents the non-linear activation function; \(W\) represents the weight matrix.

[0122] Then, the \(t\)-th level embedding vector of user \(a\) and the \((k + 1)\)-th level neighborhood influence vector of the set of neighboring users corresponding to user \(a\) are aggregated to obtain the \((k + 1)\)-th level embedding vector of user \(a\). Specifically, the aggregation of the \(k\)-th level embedding vector and the \((k + 1)\)-th level neighborhood influence vector can be carried out according to the following formula:

[0123]

[0124] Among them, \(W\) t represents the weight matrix used for \(t\)-th level social aggregation. If the social aggregation level reaches the set level threshold \(k\), then the \(k\)-th level embedding vector of user \(a\) is used as the target embedding vector of user \(a\).

[0125] After that, the resource embedding vectors of all historical preference resources corresponding to user \(a\) are weighted averaged, and the result of the weighted average is summed with the target embedding vector of user \(a\) to obtain the target feature vector of user \(a\). Specifically, the target feature vector of user \(a\) can be calculated according to the following formula:

[0126]

[0127] Among them, \(u\) a represents the target feature vector of user \(a\), \(R\) a represents the set of historical preference resources of user \(a\); \(|R|\) a represents the number of resources in the set of historical preference resources of user \(a\); \(v\) j represents the resource embedding vector of a historical preference resource in the set of historical preference resources of user \(a\).

[0128] After that, the target feature vector of user \(a\) and the resource embedding vectors of the resources in the candidate resource set are vector-connected: And the connected feature vector \(z\) obtained by the vector connection 1 is used as the input of the preference prediction model, and the preference prediction model outputs the preference weight \(r\) of user \(a\) for resource \(i\) based on this connected feature vector \(z\) 1 ai ai .

[0129] The preference prediction model includes multiple neural network layers, and the operations performed in each neural network layer can be expressed as:

[0130] \(z\) 2 \(=\sigma(W\)2 ·z 1 );(Formula 10) ......

[0131] z l-1 =σ(W l-1 ·z l-2 );(Formula 11)

[0132] r ai =W l ·z l-1 (Formula 12)

[0133] Wherein, l represents the number of layers of the neural network layer of the preference prediction model, and W l represents the weight coefficient of the l-th neural network layer.

[0134] Finally, based on the preference weights of user a for each resource, resource screening is performed to determine the target resources recommended to user a, and the target resources determined by user a.

[0135] In this embodiment, the embedding vectors of each neighborhood user at each level in the neighborhood user set of the target user are aggregated to obtain the neighborhood influence vectors at each level, and the neighborhood influence vectors at each level are gradually aggregated with the embedding vectors of the target user at each level. This process is to model the recursive social aggregation using the Graph Convolutional Networks (GCNs) to spread the global social influence of the target user. Moreover, an attention mechanism is introduced to incorporate the heterogeneous connections of each neighborhood user to the target user (different neighborhood users have different influences on the user), and the embedding vector of the target user is jointly constructed from two perspectives of social depth and social strength. Thus, the comprehensiveness and accuracy of the feature expression of the target embedding vector of the target user are ensured. Therefore, the resource recommendation based on the target embedding vector of the target user can ensure the accuracy and matching degree of the target resources determined for the target user, effectively improving the efficiency of resource recommendation.

[0136] The following introduces the apparatus embodiments of the present application, which can be used to execute the methods in the above embodiments of the present application. For the details not disclosed in the apparatus embodiments of the present application, please refer to the above method embodiments of the present application.

[0137] Figure 10 is a block diagram of a resource recommendation apparatus based on neighborhood aggregation shown according to an embodiment, as Figure 10As shown in the figure, the resource recommendation device based on neighborhood aggregation includes: an aggregation module 1010, configured to perform social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user, so as to obtain the target embedding vector of the target user; the initial embedding vector is determined according to the user attribute information of the corresponding user and the preferred resource information of the corresponding user. A target feature vector generation module 1020, configured to generate a target feature vector of the target user according to the resource embedding vector of the historical preferred resources corresponding to the target user and the target embedding vector of the target user; the resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource. A preference weight determination module 1030, configured to determine the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set. A target resource determination module 1040, configured to perform resource screening in the resource set according to the preference weights, and determine the target resources recommended to the target user.

[0138] In some embodiments of the present application, the aggregation module 1010 includes: a first acquisition unit, configured to acquire the t-level embedding vector of the target user obtained by t-level social aggregation; where t is a positive integer; and a second acquisition unit, configured to acquire the (t + 1)-level neighborhood influence vector for the (t + 1)-level social aggregation; a social aggregation unit, configured to perform (t + 1)-level social aggregation on the t-level embedding vector of the target user and the (t + 1)-level neighborhood influence vector to obtain the (t + 1)-level embedding vector of the target user; where the initial embedding vector of the target user is used for the first-level social aggregation; a target embedding vector determination unit, configured to use the (t + 1)-level embedding vector of the target user as the target embedding vector of the target user if the social aggregation level (t + 1) reaches the set level threshold; a next-level social aggregation unit, configured to continue to perform the next-level social aggregation on the (t + 1)-level embedding vector of the target user if the social aggregation level (t + 1) is less than the set level threshold.

[0139] In some embodiments of the present application, the second acquisition unit includes: a third acquisition unit, configured to acquire the t-level embedding vectors of each neighborhood user in the neighborhood set; an aggregation unit, configured to aggregate the attention coefficients of all neighborhood users in the neighborhood set and the t-level embedding vectors of all neighborhood users in the neighborhood set to obtain the (t + 1)-level neighborhood influence vector corresponding to the (t + 1)-level social aggregation; where the 1-level neighborhood influence vector corresponding to the first-level social aggregation is calculated according to the initial embedding vectors of each neighborhood user in the neighborhood set.

[0140] In some embodiments of the present application, the resource recommendation device based on neighborhood aggregation further includes: a first feature vector generation module, configured to generate a first feature vector of the target user according to the user attribute information of the target user; and a second feature vector generation module, configured to generate a second feature vector of the target user according to the preferred resource information of the target user; a first fusion module, configured to fuse the first feature vector of the target user and the second feature vector of the target user to obtain an initial embedding vector of the target user.

[0141] In some embodiments of the present application, the first fusion module includes: a first vector connection unit, configured to perform vector connection on the first feature vector of the target user and the second feature vector of the target user to obtain a user connection vector; an initial embedding vector output module, configured to input the user connection vector into a first multi-layer perceptron network, and the first multi-layer perceptron network outputs the initial embedding vector of the target user according to the user connection vector.

[0142] In some embodiments of the present application, the resource recommendation device based on neighborhood aggregation further includes: a first resource feature vector generation module, configured to generate a first resource feature vector corresponding to each resource in the resource set according to the resource attribute information of each resource in the resource set; and a second resource feature vector generation module, configured to generate a second resource feature vector corresponding to each resource in the resource set according to the associated user information of each resource in the resource set; a second fusion module, configured to fuse the first resource feature vector corresponding to the resource and the second resource feature vector corresponding to the resource to obtain a resource embedding vector corresponding to the resource.

[0143] In some embodiments of the present application, the resource recommendation device based on neighborhood aggregation further includes: a social trust degree calculation module, configured to calculate the social trust degree between the target user and each candidate user in the candidate user set according to the social interaction information of the target user; a resource preference similarity calculation module, configured to calculate the resource preference similarity between the target user and each candidate user in the candidate user set according to the preferred resource information of the target user and the preferred resource information of each candidate user in the candidate user set; a candidate user screening module, configured to screen candidate users in the candidate user set according to the social trust degree and the resource preference similarity between the target user and each candidate user in the candidate user set; an addition module, configured to add the screened candidate users to the neighborhood user set corresponding to the target user.

[0144] In some embodiments of the present application, the preference weight determination module 1030 includes: a second vector connection unit configured to perform vector connection on the target feature vector of the target user and the resource embedding vector of each resource in the resource set to obtain a connection feature vector of the target user with respect to each resource; an input unit configured to input the connection feature vector of the target user with respect to each resource into a preference prediction model; and a preference weight output unit configured to perform preference weight prediction by the preference prediction model according to the connection feature vector of the target user with respect to each resource and output the preference weights of the target user for each resource.

[0145] Figure 11 FIG. shows a schematic structural diagram of a computer system of an electronic device suitable for implementing embodiments of the present application. It should be noted that Figure 11 The illustrated computer system 1100 of the electronic device is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0146] As Figure 11 shown, the computer system 1100 includes a central processing unit (CPU) 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage section 1108 into a random access memory (RAM) 1103, such as executing the method in the above embodiment. In the RAM 1103, various programs and data required for system operation are also stored. The CPU 1101, the ROM 1102, and the RAM 1103 are connected to each other through a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0147] The following components are connected to the I / O interface 1105: an input section 1106 including a keyboard, a mouse, etc.; an output section 1107 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I / O interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed so that a computer program read from the removable medium 1111 is installed into the storage section 1108 as needed.

[0148] Specifically, according to an embodiment of the present application, the processes described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1109, and / or installed from the removable medium 1111. When the computer program is executed by a central processing unit (CPU) 1101, various functions defined in the system of the present application are executed.

[0149] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present application, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. Among them, each block in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0151] The units involved in the embodiments of the present application can be implemented in software or in hardware, and the described units can also be provided in a processor. Among them, the names of these units do not constitute a limitation to the unit itself in some cases.

[0152] On the other hand, the present application also provides a computer-readable storage medium, which can be included in the electronic device described in the above embodiments; or can exist alone without being assembled into the electronic device. The above computer-readable storage medium carries computer-readable instructions, and when the computer-readable storage instructions are executed by a processor, the methods in any of the above embodiments are implemented.

[0153] According to one aspect of the present application, an electronic device is further provided, which includes: a processor; a memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, the methods in any of the above embodiments are implemented.

[0154] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods in any of the above embodiments.

[0155] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0156] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or can be implemented by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (such as a personal computer, a server, a touch terminal, or a network device, etc.) to execute the methods according to the embodiments of the present application.

[0157] Other embodiments of the present application will be readily apparent to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of the present application and include known common general knowledge or conventional technical means in the technical field of the present application that are not disclosed in the present application.

[0158] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. A resource recommendation method based on neighborhood aggregation, characterized in that, it includes: Performing social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user; The initial embedding vector is determined according to the user attribute information of the corresponding user and the preferred resource information of the corresponding user; Generating a target feature vector of the target user according to the resource embedding vector of the historical preferred resources corresponding to the target user and the target embedding vector of the target user; The resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource; Determining the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set; Performing resource screening in the resource set according to the preference weights to determine the target resources recommended to the target user; Wherein, the determining the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set includes: Performing vector connection on the target feature vector of the target user and the resource embedding vector of each resource in the resource set to obtain the connection feature vector of the target user with respect to each resource; Inputting the connection feature vector of the target user with respect to each resource into a preference prediction model; The preference prediction model predicts the preference weights according to the connection feature vector of the target user with respect to each resource and outputs the preference weights of the target user for each resource.

2. The method according to claim 1, characterized in that, The performing social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighborhood users corresponding to the target user to obtain the target embedding vector of the target user includes: Obtaining the t-level embedding vector of the target user obtained by t-level social aggregation; where t is a positive integer; and Obtaining the (t + 1)-level neighborhood influence vector for the (t + 1)-level social aggregation; Performing (t + 1)-level social aggregation on the t-level embedding vector of the target user and the (t + 1)-level neighborhood influence vector to obtain the (t + 1)-level embedding vector of the target user; wherein, the initial embedding vector of the target user is used for the 1-level social aggregation; If the social aggregation level (t + 1) reaches the set level threshold, using the (t + 1)-level embedding vector of the target user as the target embedding vector of the target user; If the social aggregation level (t + 1) is less than the set level threshold, continuing to perform the next-level social aggregation on the (t + 1)-level embedding vector of the target user.

3. The method according to claim 2, characterized in that, The obtaining the (t + 1)-level neighborhood influence vector for the (t + 1)-level social aggregation includes: Obtaining the t-level embedding vectors of each neighborhood user in the neighborhood set; Aggregate the attention coefficients of all neighboring users in the neighborhood set and the t-level embedding vectors of all neighboring users in the neighborhood set to obtain the (t + 1)-level neighborhood influence vector corresponding to the (t + 1)-level social aggregation; among them, the 1-level neighborhood influence vector corresponding to the 1-level social aggregation is calculated based on the initial embedding vectors of each neighboring user in the neighborhood set.

4. The method according to claim 1, wherein, before the method aggregates the neighborhood influence vector of the target user through the initial embedding vector of the target user and the set of neighboring users corresponding to the target user to obtain the target embedding vector of the target user, the method further includes: generating a first feature vector of the target user according to the user attribute information of the target user; and generating a second feature vector of the target user according to the preference resource information of the target user; fusing the first feature vector of the target user and the second feature vector of the target user to obtain the initial embedding vector of the target user.

5. The method according to claim 4, wherein, the fusing the first feature vector of the target user and the second feature vector of the target user to obtain the initial embedding vector of the target user includes: performing vector connection on the first feature vector of the target user and the second feature vector of the target user to obtain a user connection vector; inputting the user connection vector into a first multi-layer perceptron network, and the first multi-layer perceptron network outputs the initial embedding vector of the target user according to the user connection vector.

6. The method according to claim 1, wherein, before the method determines the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set, the method further includes: generating a first resource feature vector corresponding to the resource according to the resource attribute information of each resource in the resource set; and generating a second resource feature vector corresponding to the resource according to the associated user information of each resource in the resource set; fusing the first resource feature vector corresponding to the resource and the second resource feature vector corresponding to the resource to obtain the resource embedding vector corresponding to the resource.

7. The method according to claim 1, wherein, before the method aggregates the neighborhood influence vector of the target user through the initial embedding vector of the target user and the set of neighboring users corresponding to the target user to obtain the target embedding vector of the target user, the method further includes: calculating the social trust degree between the target user and each candidate user in the candidate user set according to the social interaction information of the target user; calculating the resource preference similarity between the target user and each candidate user in the candidate user set according to the preference resource information of the target user and the preference resource information of each candidate user in the candidate user set; screening candidate users in the candidate user set according to the social trust degree and resource preference similarity between the target user and each candidate user in the candidate user set; Add the selected candidate users to the set of neighboring users corresponding to the target user.

8. A resource recommendation device based on neighborhood aggregation, characterized in that it includes: An aggregation module, configured to perform social aggregation on the neighborhood influence vector of the target user according to the initial embedding vector of the target user and the set of neighboring users corresponding to the target user, to obtain the target embedding vector of the target user; The initial embedding vector is determined according to the user attribute information of the corresponding user and the preference resource information of the corresponding user; A target feature vector generation module, configured to generate a target feature vector of the target user according to the resource embedding vector of the historical preference resources corresponding to the target user and the target embedding vector of the target user; The resource embedding vector is generated according to the resource attribute information of the corresponding resource and the associated user information of the corresponding resource; A preference weight determination module, configured to determine the preference weights of the target user for each resource in the resource set according to the target feature vector of the target user and the resource embedding vectors of each resource in the resource set; A target resource determination module, configured to perform resource screening in the resource set according to the preference weights, and determine the target resources recommended to the target user; wherein, the preference weight determination module includes: A second vector connection unit, configured to perform vector connection on the target feature vector of the target user and the resource embedding vector of each resource in the resource set, to obtain a connection feature vector of the target user with respect to each resource; An input unit, configured to input the connection feature vector of the target user with respect to each resource into a preference prediction model; A preference weight output unit, configured to perform preference weight prediction by the preference prediction model according to the connection feature vector of the target user with respect to each resource, and output the preference weights of the target user for each resource.

9. An electronic device, characterized in that it includes: A processor; A memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, the method described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by a processor, the method described in any one of claims 1-7 is implemented.

Images