Resource recommendation method and device and electronic equipment

Abstract

The invention provides a resource recommendation method and device and electronic equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of large models, intelligent recommendation, deep learning and the like. The specific implementation scheme is as follows: acquiring a first historical resource having an interaction behavior with a target user; acquiring a first interest point to which the first historical resource belongs; obtaining a long-tail interest point interested by the target user from the first interest point; and performing resource recommendation based on the resources under the long-tail interest points.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to technical fields such as large models, intelligent recommendations and deep learning, and especially to resource recommendation methods, devices and electronic equipment. Background Art

[0002] With the development of computer technology, the Internet can provide users with more and more network services. For example, users can browse or watch resources such as videos, articles or items through Internet platforms. In order to facilitate users to obtain information, Internet platforms can recommend resources to users through recommendation systems. The accuracy of resource recommendations is an issue that needs attention. Summary of the invention

[0003] The present disclosure provides a resource recommendation method, device and electronic device.

[0004] According to a first aspect of the present disclosure, a resource recommendation method is provided, including: obtaining a first historical resource with which a target user has an interactive behavior; obtaining a first point of interest to which the first historical resource belongs; obtaining long-tail points of interest that the target user is interested in from the first point of interest; and recommending resources based on the resources under the long-tail points of interest.

[0005] According to a second aspect of the present disclosure, a resource recommendation device is provided, including: a first acquisition module, used to acquire a first historical resource with which a target user has an interactive behavior; a second acquisition module, used to acquire a first point of interest to which the first historical resource belongs; a third acquisition module, used to acquire long-tail points of interest that the target user is interested in from the first point of interest; and a first recommendation module, used to recommend resources based on the resources under the long-tail points of interest.

[0006] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0007] at least one processor; and

[0008] a memory communicatively connected to the at least one processor; wherein,

[0009] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resource recommendation method as described in the first aspect.

[0010] 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 cause the computer to execute the resource recommendation method as described in the first aspect.

[0011] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the resource recommendation method as described in the first aspect.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a flowchart of a resource recommendation method provided according to an embodiment of the present disclosure;

[0015] Figure 2 is a flowchart of a resource recommendation method provided according to another embodiment of the present disclosure;

[0016] Figure 3 is an example diagram of a resource recommendation method provided according to an embodiment of the present disclosure;

[0017] Figure 4 is a flowchart of a resource recommendation method provided according to another embodiment of the present disclosure;

[0018] Figure 5 This is an example diagram of a resource recommendation method provided according to another embodiment of the present disclosure;

[0019] Figure 6 is a flowchart of a resource recommendation method provided according to an embodiment of the present disclosure;

[0020] Figure 7 is a schematic diagram of a resource recommendation device according to an embodiment of the present disclosure

[0021] Figure 8 It is a block diagram of an electronic device used to implement the resource recommendation method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, comply with the relevant laws and regulations, and do not violate public order and good morals.

[0024] The recommendation system in related technologies mainly uses methods such as user-item dual tower model, multi-target graph retrieval model, graph collaborative filtering model to recall resources. The model is based on the user's historical behavior data and recommends similar resources to users through similarity calculation and other methods. These methods are easily affected by system bias, resulting in over-concentration of recommendation results and failure to effectively cover the user's specific interests. For example, it is impossible to recommend resources under the long-tail interest points that the user is interested in, resulting in the recommended resources failing to meet the user's real needs, poor accuracy of resource recommendations, and poor resource recommendation effects.

[0025] Among them, interest points refer to specific areas of interest, topics or types. Long-tail interest points, that is, long-tail interest points, can refer to interest points with a narrow audience and are less common.

[0026] In order to improve the accuracy of resource recommendation and thus improve the effect of resource recommendation, the embodiments of the present disclosure provide a resource recommendation method, device and electronic device.

[0027] The resource recommendation method, device and electronic device according to the embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0028] It should be noted that the resource recommendation method of this embodiment is implemented by a resource recommendation device, which can be implemented by software and / or hardware and can be configured in an electronic device, which can include but is not limited to a terminal, a server, and the like.

[0029] Among them, the resource recommendation device can be any Internet platform that requires individualized recommendations, such as a content distribution platform, an e-commerce platform, a social media platform, or can be configured in any of the above platforms, such as a recommendation system in any of the above platforms, and the present disclosure does not impose any restrictions on this.

[0030] Figure 1 It is a flowchart of a resource recommendation method provided according to an embodiment of the present disclosure.

[0031] like Figure 1 As shown, the resource recommendation method includes:

[0032] Step 101: Acquire a first historical resource that has an interactive behavior with a target user.

[0033] The first historical resource is a resource that interacted with the target user in a certain period of time in the past. The resource may include any form of resource such as video, article, item, music, web page, etc. The number of the first historical resource may be one or more, and the present disclosure does not limit this.

[0034] Interactive behaviors may include any type of interactive behaviors such as user browsing, clicking, collecting, sharing, commenting, searching, subscribing, and forwarding.

[0035] In some embodiments, historical behavior information of the target user in the past period of time may be obtained, where the historical behavior information may include interactive behavior information such as browsing, clicking, and collecting of any resource by the user, such as information that the target user watches videos, browses articles, likes or comments on videos or articles on a web page, etc. Then, based on the historical behavior information, the first historical resource with which the target user has interactive behavior may be determined.

[0036] It should be noted that the resource recommendation device in the embodiment of the present disclosure can obtain the historical behavior information of the target user through various public, legal and compliant methods, for example, the historical behavior information of the target user can be obtained from a public data set, or the historical behavior information of the target user can be obtained from the target user after the target user authorizes it. The present disclosure does not impose any restrictions on this.

[0037] Step 102: Acquire a first point of interest to which a first historical resource belongs.

[0038] The first points of interest of different first historical resources may be the same or different. For example, assuming that the first historical resources include at least video A, video B, and article C, video A and article C may both belong to the point of interest of "electronic games", and video B may belong to the point of interest of "classical music".

[0039] Step 103: Obtain long-tail interest points that the target user is interested in from the first interest points.

[0040] Among them, the long-tail points of interest that target users are interested in are points of interest that target users are interested in but have a narrow audience and are less common.

[0041] Step 104: recommend resources based on the resources under the long-tail interest points.

[0042] In some embodiments, after obtaining the long-tail interest points that the user is interested in, the resources under the long-tail interest points can be obtained, and then the resources to be recommended can be obtained from these resources, and the resources to be recommended can be pushed to the target user.

[0043] The resource recommendation method provided by the embodiment of the present disclosure obtains the first historical resource with which the target user has an interactive behavior; obtains the first interest point to which the first historical resource belongs; obtains the long-tail interest point that the target user is interested in from the first interest point; and recommends resources based on the resources under the long-tail interest point. In this way, the long-tail interest point that the target user is interested in can be obtained, and then the resources under the long-tail interest point can be recommended to the target user, thereby solving the problem that the recommendation system is too concentrated in the recommendation results due to the influence of system deviation, and the resources under the long-tail interest point that the user is interested in cannot be recommended to the user, thereby meeting the user's demand for resources under the long-tail interest point, improving the accuracy of resource recommendation, optimizing the resource recommendation effect, and improving user satisfaction.

[0044] In order to clearly explain the process of obtaining the first interest point belonging to the first historical resource and obtaining the long-tail interest points of interest of the target user from the first interest point, as well as the process of recommending resources based on the resources under the long-tail interest points, the embodiment of the present disclosure also provides a resource recommendation method.

[0045] Figure 2 It is a flowchart of a resource recommendation method provided according to another embodiment of the present disclosure.

[0046] like Figure 2 As shown, the resource recommendation method includes:

[0047] Step 201: Acquire a first historical resource that has an interactive behavior with a target user.

[0048] The specific implementation process and principle of step 201 may refer to the description of other embodiments and will not be repeated here.

[0049] Step 202: based on a first corresponding relationship between a first preset resource and a quantization code, obtain a target quantization code corresponding to a first historical resource, wherein the target quantization code represents a point of interest to which the first historical resource belongs.

[0050] The first preset resource is a preset resource that can be obtained in a variety of ways, such as obtaining historical behavior information of any number of users, and based on the historical behavior information, obtaining resources that have interactive behaviors with these users, and using these resources as the first preset resources.

[0051] It should be noted that the resource recommendation device in the embodiment of the present disclosure can obtain the historical behavior information of one or more users through various public, legal and compliant methods, for example, it can obtain the historical behavior information of the user from a public data set, or it can obtain the historical behavior information from the user after the user's authorization. The present disclosure does not impose any restrictions on this.

[0052] The quantization code corresponding to the first preset resource is obtained by performing a quantization coding process on the first preset resource, and the obtained quantization code can represent the point of interest to which the corresponding first preset resource belongs. The quantization coding process is a process of converting continuous, high-dimensional data of the first preset resource into a discrete, low-dimensional representation, and in this process, important information of the original data is retained as much as possible.

[0053] In some embodiments, a first correspondence between a first preset resource and a quantization code may be generated in advance and saved. For any first historical resource, a first preset resource identical or similar to the first historical resource may be queried based on the first correspondence, and then the quantization code corresponding to the queried first preset resource may be determined as the target quantization code corresponding to the first historical resource.

[0054] Step 203: Determine a first interest point based on the interest point represented by the target quantization code.

[0055] In some embodiments, for any first preset resource, the point of interest represented by the corresponding target quantization code may be used as the first point of interest to which the first preset resource belongs.

[0056] Therefore, by pre-generating a first correspondence between the first preset resource and the quantization code, and then directly based on the first correspondence, obtaining the target quantization code corresponding to the first historical resource, wherein the target quantization code represents the point of interest belonging to the first historical resource, and then determining the first point of interest based on the point of interest represented by the target quantization code, the first point of interest belonging to any first preset resource can be quickly determined.

[0057] In some embodiments, the number of the first preset resources is at least two, and the first corresponding relationship between each first preset resource and the quantization code can be determined in the following manner:

[0058] Based on the similarity between the features of the first preset resources, clustering the first preset resources to obtain a first cluster to which the first preset resources belong, wherein the first cluster represents the type of the point of interest to which the first preset resource belongs;

[0059] Determine a first distance between the first preset resource and the cluster center of the first cluster to which it belongs, and cluster the first distance to obtain a second cluster to which the first distance belongs, wherein the second cluster represents a subtype to which the first preset resource belongs under the type of the point of interest to which it belongs;

[0060] Obtaining a first code of a cluster center of a first cluster to which the first preset resource belongs and a second code of a cluster center of a second cluster to which the corresponding first distance belongs, and determining a quantized code representing a point of interest to which the first preset resource belongs based on the first code and the second code;

[0061] A first corresponding relationship is determined based on the quantization code corresponding to the first preset resource.

[0062] It is understandable that based on the similarity between the features of at least two first preset resources, these first preset resources can be clustered to obtain the first cluster to which any first preset resource belongs, wherein the number of first clusters obtained by clustering can be one or at least two, and each first cluster represents the type of interest point to which the first preset resource belongs. The clustering method can adopt K-means clustering or other clustering methods, and the present disclosure does not limit this.

[0063] Furthermore, for any first preset resource among the at least two first preset resources mentioned above, the first distance between the first preset resource and the cluster center of the first cluster to which it belongs can be determined, thereby obtaining at least two first distances. By clustering the obtained first distances, the second cluster to which any first distance belongs can be obtained, wherein the number of second clusters obtained by clustering can be one or at least two, and each second cluster represents the subtype to which the first preset resource corresponding to the first distance belongs under the type of the corresponding point of interest. The clustering method can adopt K-means clustering or other clustering methods, and the present disclosure does not limit this.

[0064] The features of different types of first preset resources may be different. For example, taking the first preset resource as an article, its features may include themes, keywords, content, length, etc.; taking the first preset resource as a video, its features may include resolution, duration, bit rate, category, etc.; taking the first preset resource as an image, its features may include resolution, format, color space, size, shooting location, etc.

[0065] In order to classify and describe the points of interest, the points of interest may be divided into different types, and the types of the points of interest may be further divided to obtain subtypes of the points of interest.

[0066] Among them, for a certain first preset resource, the first cluster to which it belongs can, for example, indicate that the type of the point of interest to which the first preset resource belongs is "entertainment", and the second cluster to which the first distance corresponding to the first preset resource belongs can, for example, indicate that the first preset resource belongs to the subtype "movie" under the point of interest type "entertainment".

[0067] It should be noted that when clustering the obtained first distances, the first distances corresponding to the first preset resources in each first cluster can be clustered individually, or the first distances corresponding to the first preset resources in each first cluster can be merged into a set and the first distances in the set can be clustered. Other methods can also be used for clustering. The clustering method can be set as needed, and the present disclosure does not impose any restrictions on this.

[0068] Furthermore, for any first cluster, the first code of the cluster center of the first cluster can be obtained, wherein the first code is the index of the corresponding cluster center of the first cluster, and the indexes of the cluster centers of different first clusters are different; for any second cluster, the second code of the cluster center of the second cluster can be obtained, wherein the second code is the index of the corresponding cluster center of the second cluster, and the indexes of the cluster centers of different second clusters are different. Furthermore, for any first preset resource, the quantization code representing the point of interest to which the first preset resource belongs can be determined based on the first code of the cluster center of the first cluster to which it belongs, and the second code of the cluster center of the second cluster to which the corresponding first distance belongs. Based on the quantization code corresponding to any first preset resource, the first corresponding relationship between the first preset resource and the quantization code can be determined.

[0069] Among them, the method of obtaining the corresponding quantization code according to the first code and the second code can be set as needed. For example, assuming that the number of first clusters is 128, the cluster centers of the 128 first clusters can be numbered, such as numbered from 1 to 128, and the number is used as the index of the corresponding cluster center, that is, the first code. Similarly, assuming that the number of second clusters is 128, the cluster centers of the 128 second clusters can be numbered, such as numbered from 1 to 128, and the number is used as the index of the corresponding cluster center, that is, the second code. The method of obtaining the corresponding quantization code according to the first code and the second code can be: Z=X*P+Y. Among them, X represents the first code, Y represents the second code, and Z represents the obtained quantization code. Among them, P is a preset coefficient, which can be determined according to the number of first clusters. For example, in this example, P can be set to 1000 according to the number of bits of the first cluster 128.

[0070] It should be noted that the quantization codes corresponding to different first preset resources may be the same or different, and accordingly, the points of interest to which different first preset resources belong may be the same or different.

[0071] In some embodiments, for any first preset resource, a feature vector of the first preset resource can be obtained, and the feature vector represents the feature of the first preset resource. Furthermore, when clustering at least two first preset resources, clustering can be performed based on the similarity between the feature vectors of these first preset resources. Accordingly, the first distance corresponding to any first preset resource can be the distance (also referred to as a residual) between the feature vector of the first preset resource and the feature vector of the cluster center of the first cluster to which it belongs. The feature vector of any resource can also be referred to as a resource feature vector.

[0072] Therefore, by only clustering at least two first preset resources and clustering the first distances corresponding to at least two first preset resources, the quantization coding process of the first preset resources can be quickly completed based on the clustering results, and the quantization coding corresponding to the first preset resources can be obtained, and then the first corresponding relationship between the first preset resources and the quantization coding can be efficiently determined.

[0073] In some embodiments, the resource recommendation method may further include:

[0074] Based on the similarity between the features of the second preset resource and the cluster center of the first cluster, determining from the first cluster a third cluster to which the second preset resource belongs, wherein the third cluster represents the type of the point of interest to which the second preset resource belongs;

[0075] Determine a second distance between the second preset resource and the cluster center of the third cluster to which it belongs, and determine a fourth cluster to which the second distance belongs from the second cluster based on the similarity between the second distance and the cluster center of the second cluster, wherein the fourth cluster represents a subtype to which the second preset resource belongs under the type of the point of interest to which it belongs;

[0076] Acquire a third code of the cluster center of the third cluster and a fourth code of the cluster center of the fourth cluster, and determine a quantized code representing the point of interest to which the second preset resource belongs based on the third code and the fourth code;

[0077] The first corresponding relationship is updated based on the quantization code corresponding to the second preset resource.

[0078] The second preset resource is a preset resource, and the acquisition method can refer to the acquisition method of the first preset resource, which will not be repeated here. The number of the second preset resources can be one or at least two, which is not limited in the present disclosure.

[0079] Among them, the characteristics of different types of second preset resources may be different.

[0080] It can be understood that, for any second preset resource, the third cluster to which the second preset resource belongs can be determined from each first cluster based on the similarity between the features of the second preset resource and the cluster centers of each first cluster, and the third cluster can represent the type of the interest point to which the second preset resource belongs. Further, the second distance between the second preset resource and the cluster center of the third cluster to which it belongs can be determined, and based on the similarity between the second distance and the cluster centers of each second cluster, the fourth cluster to which the second distance belongs can be determined from each second cluster, wherein the fourth cluster represents the subtype to which the second preset resource belongs under the type of the interest point to which it belongs. Further, for any second preset resource, the quantization code representing the interest point to which the second preset resource belongs can be determined based on the third code of the cluster center of the third cluster to which it belongs, and the fourth code of the cluster center of the fourth cluster to which the corresponding second distance belongs. Wherein, the method of determining the corresponding quantization code based on the third code and the fourth code is the same as the method of determining the corresponding quantization code based on the first code and the second code.

[0081] Among them, for any second preset resource, since the third cluster to which it belongs is determined from the first cluster, and the fourth cluster to which the corresponding second distance belongs is determined from the second cluster, the third code of the cluster center of the third cluster is the first code of the cluster center of a certain first cluster, and the fourth code of the cluster center of the fourth cluster is the second code of the cluster center of a certain second cluster. Then, the quantization code determined according to the third code and the fourth code may be the same as the quantization code corresponding to a certain first preset resource that has been determined, then, the second preset resource can be used as the first preset resource corresponding to the quantization code to supplement the first preset resource corresponding to the quantization code to obtain an updated first corresponding relationship.

[0082] In some embodiments, for any second preset resource, a feature vector of the second preset resource can be obtained, and the feature vector represents the feature of the second preset resource. Furthermore, based on the similarity between the feature vector of the second preset resource and the feature vector of the cluster center of any first cluster, the similarity between the features of the second preset resource and the cluster center of the first cluster can be determined. Correspondingly, the second distance between the second preset resource and the cluster center of the third cluster to which it belongs can be the distance (also referred to as a residual) between the feature vector of the second preset resource and the feature vector of the cluster center of the third cluster to which it belongs. Among them, the feature vector of any resource can also be referred to as a resource feature vector.

[0083] Therefore, based on the similarity between the characteristics of the second preset resource and the cluster center of the first cluster, the third cluster to which the second preset resource belongs can be quickly determined from the first cluster, and based on the similarity between the second distance corresponding to the second preset resource and the cluster center of the second cluster, the fourth cluster to which the second distance belongs can be quickly determined from the second cluster, and then based on the third code of the cluster center of the third cluster and the fourth code of the cluster center of the fourth cluster, the quantization code representing the interest point to which the second preset resource belongs can be quickly determined, thereby achieving efficient updating of the first corresponding relationship.

[0084] In some embodiments, the cluster center of the third cluster can also be updated based on the second preset resource and the first preset resource included in the third cluster to which it belongs, and the cluster center of the fourth cluster can be updated based on the second distance corresponding to the second preset resource and the first distance included in the fourth cluster to which it belongs, so that the cluster center of the third cluster and the cluster center of the fourth cluster are more accurate, so that when the quantization code corresponding to other preset resources is subsequently determined based on the cluster center of the third cluster and the cluster center of the fourth cluster, the accuracy of the determined quantization code can be improved.

[0085] In some embodiments, updating the cluster center of the third cluster based on the second preset resource and the first preset resource included in the third cluster to which it belongs may include: obtaining a feature vector of the second preset resource and a feature vector of the first preset resource included in the third cluster to which it belongs, and performing weighted averaging on the two types of feature vectors, and using the obtained feature vector as the feature vector of the cluster center of the third cluster.

[0086] Correspondingly, the second distance corresponding to the second preset resource is the distance between the feature vector of the second preset resource and the feature vector of the cluster center of the third cluster to which it belongs, and the first distance included in the fourth cluster to which the second distance belongs is the distance between the feature vector of the first preset resource and the feature vector of the cluster center of the second cluster to which it belongs. Updating the cluster center of the fourth cluster based on the second distance corresponding to the second preset resource and the first distance included in the fourth cluster to which it belongs may include: taking a weighted average of the first distance and the second distance, and using the obtained feature vector as the feature vector of the cluster center of the fourth cluster.

[0087] As a possible implementation, assuming that the resource recommendation device obtains feature vectors of M1 first preset resources, the M1 feature vectors can be clustered based on the similarity between the M1 feature vectors to obtain the first cluster to which any first preset resource belongs, wherein M1 is an integer greater than 1. In addition, the first distance between the M1 feature vectors and the feature vector of the cluster center of the first cluster to which they belong can be obtained, and the obtained M1 first distances can be clustered to obtain the second cluster to which any first distance belongs. Assuming that the number of first clusters and second clusters is N, and N is an integer greater than or equal to 1, a first-level codebook and a second-level codebook can be generated, wherein the first-level codebook includes feature vectors corresponding to the cluster centers of the N first clusters, and the second-level codebook includes feature vectors corresponding to the cluster centers of the N second clusters. In addition, the first codes corresponding to the cluster centers of the N first clusters and the second codes corresponding to the cluster centers of the N second clusters can be obtained.

[0088] For any first preset resource, the first code of the cluster center of the first cluster to which it belongs can be obtained, and the second code of the cluster center of the second cluster to which the first distance belongs can be obtained, and according to Z=X*P+Y, the quantization code corresponding to the first preset resource can be obtained. Thus, N*N quantization codes and the first preset resource corresponding to each quantization code can be obtained, thereby obtaining the corresponding relationship between the first preset resource and the quantization code.

[0089] Further, it is assumed that the resource recommendation device obtains the feature vectors of M2 second preset resources, where M2 is an integer greater than or equal to 1. Then for any second preset resource, based on the similarity between the feature vector of the second preset resource and the feature vector of the cluster center of the N first clusters, the third cluster to which the second preset resource belongs can be determined from the N first clusters, and the second distance between the feature vector of the second preset resource and the feature vector of the cluster center of the third cluster is determined, and based on the similarity between the second distance and the feature vector of the cluster center of the N second clusters, the fourth cluster to which the second preset resource belongs is determined from the N second clusters. Assuming that the third code of the cluster center of the third cluster is x and the fourth code of the cluster center of the fourth cluster is y, it can be determined that the quantization code corresponding to the second preset resource is x*P+y, and then the second preset resource can be used as the first preset resource corresponding to the quantization code x*P+y to supplement the first preset resource corresponding to the quantization code x*P+y.

[0090] In addition, after determining the third cluster to which M2 second preset resources belong and the fourth cluster to which the corresponding second distance belongs, assuming that a third cluster includes M3 second preset resources, where M3 is greater than or equal to 1 and less than or equal to M2, the characteristic vector of the cluster center of the third cluster can be updated based on the characteristic vector of the M3 second preset resources, and the cluster center of the fourth cluster can be updated based on the second distance corresponding to the M3 second preset resources.

[0091] Step 204: Obtain long-tail interest points that the target user is interested in from the first interest points.

[0092] In some embodiments, step 204 may be implemented by:

[0093] Obtain the number of clicks on resources under at least two preset points of interest;

[0094] Determine a second point of interest from the preset points of interest based on the number of clicks, wherein the number of clicks on resources under the second point of interest is lower than the number of clicks on resources under other points of interest among the preset points of interest except the second point of interest;

[0095] The interest points belonging to the second interest points in the first interest points are determined as long-tail interest points.

[0096] The preset interest points are pre-set interest points, and the acquisition method thereof can be set as required. For example, the preset interest points can include the interest points represented by the quantized codes in the above first corresponding relationship.

[0097] The number of the second points of interest may be set as needed, for example, 30, 50, etc.

[0098] As a possible implementation method, the preset points of interest may be sorted in descending order of the number of clicks, and a set number of points of interest that are sorted later may be determined as the second points of interest.

[0099] For any first point of interest, if the second points of interest include the first point of interest, the first point of interest may be determined as a long-tail point of interest that is of interest to the target user.

[0100] Therefore, the long-tail interest points that the target user is interested in in the first interest point can be determined quickly and accurately according to the number of clicks on the resources under the preset interest points.

[0101] Step 205: Determine the target user's preference for resources under the long-tail interest points through the recommendation model.

[0102] Among them, the recommendation model is any neural network model or other machine model that has been trained and can determine the preference level of any user for any resource, such as a large model, and the present disclosure does not impose any restrictions on this.

[0103] The input of the recommendation model may include the feature information of any user and the feature information of any resource, and the output of the recommendation model may include the user's preference for the resource. The preference degree can be understood as the user's liking, interest or satisfaction with the resource. The higher the preference degree, the higher the user's liking, interest or satisfaction with the resource. The preference degree can be represented by a score.

[0104] In some embodiments, step 205 may be implemented by:

[0105] Obtaining first characteristic information of target users and second characteristic information of resources under long-tail interest points;

[0106] Inputting the first feature information into a user network for generating a user feature vector in a recommendation model to obtain a user feature vector of a target user;

[0107] Inputting the second feature information into a resource network used to generate a resource feature vector in a recommendation model, and obtaining a resource feature vector of resources under the long-tail interest point;

[0108] Based on the obtained user feature vector and resource feature vector, the preference degree is determined.

[0109] The first characteristic information may include the characteristic information of the target user and the historical behavior information of the target user. It should be noted that the resource recommendation device in the embodiment of the present disclosure may obtain the first characteristic information of the target user in various public, legal and compliant ways, such as obtaining the first characteristic information of the target user from a public data set, or obtaining the first characteristic information of the target user from the target user after authorization by the target user, and the present disclosure does not limit this.

[0110] The second characteristic information of different types of resources may be different. For example, if the resource is an article, its second characteristic information may include information such as subject, keyword, content, length, etc.; if the resource is a video, its second characteristic information may include information such as resolution, duration, bit rate, category, etc.; if the resource is an image, its second characteristic information may include information such as resolution, format, color space, size, shooting location, etc.

[0111] Among them, the user feature vector, that is, the feature vector of the user, can represent the characteristics of the user. The resource feature vector, that is, the feature vector of the resource, can represent the characteristics of the resource.

[0112] In an embodiment of the present disclosure, the recommendation model may include a user network for generating a user feature vector, and a resource network for generating a resource feature vector. The user network may be a Transformer-Encoder network, or any other network capable of generating a user feature vector, which is not limited by the present disclosure. The resource network may be a network for generating a resource feature vector in a U2I (User-to-Item) model, or any other network capable of generating a resource feature vector, which is not limited by the present disclosure.

[0113] Among them, when the user network is a Transformer-Encoder network, each feature of the user can be used as a token (word unit), and the dependency relationship between each feature can be captured through the Transformer-Encoder network, so as to more fully represent the user's characteristics and have a more complete understanding of the user, laying the foundation for the subsequent accurate determination of the target user's preference for any resource based on the user feature vector of the target user.

[0114] In the disclosed embodiment, for any resource under the long-tail interest point, the preference of the target user for the resource can be obtained through multi-objective scoring based on the user feature vector of the target user and the resource feature vector of the resource.

[0115] Among them, multi-objective scoring is to evaluate the target user's preference for resources based on multiple different objectives or standards. Taking the resources such as videos, audios, or articles as an example, the objectives or standards may include: the relevance between the target user and the resource, the probability that the target user has browsed or watched the resource, etc.

[0116] Therefore, by adopting the user network in the recommendation model, based on the first feature information of the target user, the user feature vector of the target user is determined, and by adopting the resource network in the recommendation model, based on the second feature information of the resources under the long-tail interest points of the target user, the resource feature vector of the resource is determined, and then the target user's preference for the resource can be accurately determined based on the user feature vector and the resource feature vector.

[0117] In some embodiments, the recommendation model can be trained in the following manner:

[0118] Acquire training samples, wherein any training sample includes characteristic information of a sample user and characteristic information of a sample resource associated with the sample user, and the training sample is labeled with the degree of preference of the sample user for the sample resource. The characteristic information of the sample user may include characteristic information of the sample user itself and information on the interactive behavior of the sample user with respect to the sample resource;

[0119] Input the characteristic information of the sample user in the training sample into the initial user network to obtain the user characteristic vector of the sample user;

[0120] Inputting the characteristic information of the sample resources in the training sample into the initial resource network to obtain the resource characteristic vector of the sample resource;

[0121] Based on the user feature vector of the sample user and the resource feature vector of the sample resource, obtaining the predicted preference degree of the sample user for the sample resource;

[0122] The loss is calculated based on the predicted preference of sample users for sample resources and the labeled preference, and the model parameters of the initial user network and resource network are adjusted based on the calculated loss to obtain the trained user network and resource network.

[0123] Step 206 , based on the degree of preference, determine the target resource from the resources under the long-tail interest point, and recommend the target resource to the target user.

[0124] In some embodiments, the resources under the long-tail interest points that the user is interested in can be sorted in order from high to low preference, and a set number of resources ranked first can be determined as target resources, and the target resources can be recommended to the target user.

[0125] Therefore, by adopting the recommendation model to determine the target user's preference for resources under the long-tail interest points, based on the preference degree, the target resources are determined from the resources under the long-tail interest points, and the target resources are recommended to the target user, thereby recommending the resources under the long-tail interest points of interest to the target user, thereby meeting the user's demand for resources under the long-tail interest points, improving the accuracy of resource recommendation, optimizing the resource recommendation effect, and improving user satisfaction.

[0126] In practical applications, the target resources among the resources under the long-tail interest points of interest of the user determined by the embodiments of the present disclosure can be used as a supplement to the resources recommended by the recommendation system in the related technology, thereby improving the recommendation effect of the recommendation system on the resources under the long-tail interest points.

[0127] In some embodiments, the resource recommendation device may recommend resources to the target user based on resources under the long-tail interest points when it is determined that the resources recommended to the second historical user do not include resources under the long-tail interest points that the second historical user is interested in.

[0128] The second historical user may be any user who has made resource recommendations.

[0129] Among them, the long-tail interest points of the second historical user are obtained based on the historical behavior information of the second historical user on the third historical resource. The third historical resource is a resource obtained from a resource queue and recommended to the second historical user. The resource queue includes resources randomly obtained from a resource library.

[0130] It should be noted that the resource recommendation device in the embodiment of the present disclosure can obtain the historical behavior information of the second historical user on the third historical resource through various public, legal and compliant methods. For example, the historical behavior information of the second historical user on the third historical resource can be obtained from a public data set, or the historical behavior information of the second historical user on the third historical resource can be obtained from the second historical user after authorization. The present disclosure does not impose any restrictions on this.

[0131] The resource queue may also be called an unbiased forced insertion queue or other queues, and the present disclosure does not limit this.

[0132] As a possible implementation, the resource recommendation device can be a recommendation system in an Internet platform. The resource recommendation device can first randomly obtain some resources from the resource library and save them to a resource queue, and push the resources in the resource queue to the front end of the Internet platform. The front end can obtain some resources, namely the third historical resources, from them and recommend them to the second historical user. According to the second historical user's interactive behavior such as clicking or viewing the recommended third historical resource, a log including these historical behavior information can be generated. Statistical analysis based on the log can determine the long-tail interest points of the second historical user.

[0133] The front end may randomly obtain the third historical resource from the resources in the resource queue and push it to the second historical user, or the front end may obtain the third historical resource from the resource queue using a specific strategy based on the characteristic information of the resources in the resource queue and push it to the second historical user. The specific strategy may include, for example, obtaining resources that match the age group of the second historical user from the resources in the resource queue.

[0134] Combine the following Figure 3 Taking the resource recommendation device as a recommendation system in an Internet platform as an example, the relationship between the resource queue 301, the user (including the second historical user, the target user and other users) 302, the recommendation system 303, and the resource library 304 is exemplarily explained.

[0135] refer to Figure 3, the resource queue 301 is independent of the recommendation system 303 and does not use the recommendation method in the recommendation system 303. The resource queue 301 can randomly obtain resources from the resource library 304, and adopt a specific strategy or randomly recommend some resources in the resource queue 301 to a user in the user 302, so as to obtain the historical behavior information of the user on the recommended resources. Based on this historical behavior information, the recommendation system 303 can determine the long-tail interest points that the user is interested in, and determine whether the resources recommended to the user by the recommendation system 303 include the resources under the long-tail interest points that the user is interested in. If not, the resource recommendation method of the recommendation system 303 can be optimized, so that for any user in the user 302, when making resource recommendations later, the long-tail interest points that the user is interested in can be determined in the manner shown in steps 201-204, and resources can be recommended to the user based on the resources under the long-tail interest points in the manner shown in steps 205-206. Among them, Figure 3 The arrow from the resource queue 301 to the recommendation system 303 indicates that the historical behavior information used to optimize the recommendation system 303 is obtained based on the resources in the resource queue 301 .

[0136] Therefore, by obtaining the third historical resource from the resource queue and recommending it to the second historical user, wherein the resource queue includes resources randomly obtained from the resource library, the long-tail interest points that the second historical user is actually interested in can be determined based on the historical behavior information of the second historical user on the third historical resource without being affected by system deviations. Further, when the resource recommendation device determines that the resources recommended to the second historical user do not include the resources under these long-tail interest points, it can be determined that there is a problem with the resource recommendation device in recommending resources for the long-tail interest points. Therefore, when subsequently recommending resources to the target user, the long-tail interest points that the target user is interested in can be determined, and based on the resources under the long-tail interest points, resource recommendations can be made to the target user to overcome this problem and improve the accuracy of resource recommendations and the resource recommendation effect.

[0137] Among them, the specific implementation forms of steps 201-206 and similar steps in other embodiments can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0138] In an embodiment of the present disclosure, a first historical resource with which a target user has an interactive behavior is obtained, and based on a first corresponding relationship between a first preset resource and a quantization code, a target quantization code corresponding to the first historical resource is obtained, wherein the target quantization code represents the interest point to which the first historical resource belongs, and based on the interest point represented by the target quantization code, a first interest point is determined, and long-tail interest points of interest to the target user are obtained from the first interest point. Through a recommendation model, the target user's preference for resources under the long-tail interest point is determined, and based on the preference degree, a target resource is determined from the resources under the long-tail interest point, and the target resource is recommended to the target user, thereby improving the accuracy of resource recommendation, optimizing the resource recommendation effect, and enhancing user satisfaction.

[0139] Through the above analysis, it can be known that the embodiments of the present disclosure can obtain the long-tail points of interest that the target user is interested in, and then recommend the resources under the long-tail points of interest to the target user, so as to improve the accuracy of resource recommendation. In some embodiments, it may also be the case that the resources under the long-term points of interest that the user is interested in cannot be recommended to the user due to system deviations. In order to further improve the accuracy of resource recommendation, the long-term points of interest that the target user is interested in can also be obtained, and then resource recommendations are made based on the resources under the long-tail points of interest that the target user is interested in and the resources under the long-term points of interest. Among them, long-term points of interest, that is, long-term points of interest, can refer to points of interest that the user is continuously interested in within a certain period of time. In view of the above situation, combined with Figure 4 , further illustrating the resource recommendation method provided in the embodiment of the present disclosure.

[0140] Figure 4 It is a flowchart of a resource recommendation method provided according to another embodiment of the present disclosure.

[0141] like Figure 4 As shown, the resource recommendation method includes:

[0142] Step 401: Acquire a first historical resource that has an interactive behavior with a target user.

[0143] The first historical resource may be determined based on historical behavior information of the target user within a certain period of time.

[0144] Step 402: Acquire a first point of interest to which a first historical resource belongs.

[0145] The specific implementation process and principle of steps 401-402 may refer to other embodiments and will not be described in detail here.

[0146] Step 403: Acquire long-tail interest points and long-term interest points that the target user is interested in from the first interest points.

[0147] Among them, the method of obtaining the long-tail interest points that the target user is interested in from the first interest point can refer to other embodiments and will not be repeated here.

[0148] In some embodiments, the long-term interest point that the target user is interested in can be obtained from the first interest point in the following manner:

[0149] determining a quantity of a first historical resource under a first point of interest;

[0150] When the number is higher than the number threshold, the first point of interest is determined as a long-term point of interest.

[0151] The quantity threshold can be set as needed.

[0152] It can be understood that the first points of interest belonging to different first historical resources can be the same, so that for any first point of interest, the number of first historical resources under the first point of interest can be determined. When the number is higher than the quantity threshold, it can be determined that the target user continues to be interested in the first point of interest within a certain period of time, and thus the first point of interest can be determined as a long-term point of interest of the target user.

[0153] Therefore, the long-term interest points that the target user is interested in among the first interest points can be determined quickly and accurately according to the number of the first historical resources under the first interest point.

[0154] Step 404 , recommending resources based on the resources under the long-tail interest points and the resources under the long-term interest points.

[0155] In some embodiments, a recommendation model can be used to determine the target user's preference for resources under long-tail points of interest and the target user's preference for resources under long-term points of interest, and based on the preference, determine the target resource from the resources under the long-tail points of interest and the resources under the long-term points of interest, and recommend the target resource to the target user.

[0156] Among them, the specific implementation forms of the steps in step 404 that are the same or similar to those in other embodiments can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0157] In some embodiments, the resource recommendation device may recommend resources to the target user based on resources under long-tail points of interest and resources under long-term points of interest when it is determined that the resources recommended to the second historical user do not include resources under long-tail points of interest and long-term points of interest that the second historical user is interested in.

[0158] The resource recommendation method provided by the disclosed embodiment can obtain the long-tail interest points and long-term interest points that the target user is interested in, and then recommend the resources under the long-tail interest points and the resources under the long-term interest points to the target user, thereby solving the problem that the recommendation system is too concentrated in the recommendation results due to the influence of system deviation, and is unable to recommend the resources under the long-tail interest points and long-term interest points that the user is interested in. This can meet the user's demand for resources under the long-tail interest points and long-term interest points, improve the accuracy and diversity of resource recommendations, optimize the resource recommendation effect, and improve user satisfaction.

[0159] The following takes the resource recommendation device as a recommendation system in the Internet platform as an example. Figure 5 , the resource recommendation method provided in the embodiment of the present disclosure is exemplarily described. Figure 5 This is an example diagram of a resource recommendation method provided according to another embodiment of the present disclosure.

[0160] The recommendation system may include a recommendation model, which includes a user network 502 for generating a user feature vector and a resource network 503 for generating a resource feature vector. The user network 502 may be a Transformer-Encoder network.

[0161] refer to Figure 5 ,The resource recommendation method can include a model training phase and an ,online recall phase.

[0162] In the model training stage, multiple rounds of iterative training can be performed to obtain the trained user network 502 and resource network 503. In each round of iterative training, multiple training samples 501 can be obtained, and any training sample 501 includes the characteristic information of the sample user and the characteristic information of the sample resources associated with the sample user, and the training sample is labeled with the preference degree of the sample user for the sample resource. By inputting the characteristic information of the sample user in any training sample 501 into the initial user network 502, the user characteristic vector 504 of the sample user can be obtained. By inputting the characteristic information of the sample resource in any training sample 501 into the initial resource network 503, the resource characteristic vector 505 of the sample resource can be obtained. Then, multi-objective scoring can be performed based on the user characteristic vector of the sample user and the resource characteristic vector of the sample resource to obtain the predicted preference degree of the sample user for the sample resource. Furthermore, loss calculation can be performed based on the predicted preference degree of sample users for sample resources and the marked preference degree, and the model parameters of the initial user network 502 and resource network 503 can be adjusted based on the calculated loss to obtain the trained user network 502 and resource network 503.

[0163] In addition, a cluster coding module 506 may be added to the recommendation model. The cluster coding module 506 may include RQ-VAE (Residual Quantized Variational Autoencoder), which may be used to perform a quantization coding process during the training of the user network 502 and the resource network 503, thereby generating a first corresponding relationship between the first preset resource and the quantization coding.

[0164] In which, in any round of training, the sample resources in any training sample can be used as the first preset resource in the above embodiment, the input of the clustering coding module 506 can include the resource feature vector 505 of the sample resource output by the resource network 503, and the output of the clustering coding module 506 can include the quantization code 511 corresponding to the sample resource, so that based on the sample resource and the corresponding quantization code, the first corresponding relationship between the first preset resource and the quantization code can be obtained.

[0165] The quantization encoding process performed by the cluster encoding module 506 may be as follows:

[0166] In any round of training, for the training samples used in this round of training, the resource feature vectors 505 of the sample resources (i.e., the first preset resources) in these training samples are obtained, and these resource feature vectors 505 are clustered to obtain the first cluster to which any sample resource belongs, and the first distance between these resource feature vectors 505 and the feature vector of the cluster center of the first cluster to which they belong is obtained, and the obtained first distance is clustered to obtain the second cluster to which any first distance belongs, and then the first code table 507 and the second code table 508 are generated. Among them, taking the example that the number of first clusters and second clusters are both 128, the first code table includes feature vectors corresponding to the cluster centers of 128 first clusters, and the second code table includes feature vectors corresponding to the cluster centers of 128 second clusters. And the first codes corresponding to the cluster centers of the 128 first clusters can be obtained ( Figure 5 1 to 128 are used as examples) and the second codes corresponding to the cluster centers of the 128 second clusters ( Figure 5 1 to 128 are used as examples for illustration). Further, for any training sample among these training samples, the first code corresponding to the cluster center of the first cluster to which the sample resource belongs can be obtained, and the second code of the cluster center of the second cluster to which the corresponding first distance belongs can be obtained, and then the quantization code 511 corresponding to the sample resource is obtained based on the first code and the second code. Thus, the first corresponding relationship between the sample resource (i.e., the first preset resource) and the quantization code can be obtained.

[0167] Assuming that in the next round of training, multiple training samples are continued to be used to train the user network 502 and the resource network 503, the training samples used in this round of training can be used as the second preset resource in the above embodiment. For the training samples used in this round of training, the resource feature vector 505 of the sample resource (i.e., the second preset resource) in any training sample can be obtained, and based on the similarity between the resource feature vector 505 and the 128 feature vectors in the first-level code table, the third cluster to which the sample resource belongs is determined from the 128 first clusters, and the second distance 510 between the resource feature vector 505 and the feature vector 509 of the cluster center of the third cluster is determined, and based on the similarity between the second distance 510 and the 128 feature vectors in the second-level code table, the fourth cluster to which the second preset resource belongs is determined from the 128 second clusters. Assuming that the third code of the cluster center of the third cluster is x, and the fourth code of the cluster center of the fourth cluster is y, the quantization code corresponding to the sample resource can be determined based on x and y ( Figure 5 Taking the quantization code xy as an example for illustration), the sample resource can be used as the first preset resource corresponding to the quantization code xy to supplement the first preset resource corresponding to the quantization code xy.

[0168] In addition, after determining the third cluster to which the above-mentioned sample resource belongs and the fourth cluster to which its corresponding second distance belongs, the characteristic vector of the cluster center of the third cluster can be updated based on the resource characteristic vector of the sample resource, and the cluster center of the fourth cluster can be updated based on the second distance corresponding to the sample resource.

[0169] Thus, a first corresponding relationship 514 between the first preset resource and the quantization code can be obtained, and then the first corresponding relationship 514 can be stored.

[0170] In the online recall phase, for any target user 512, the first historical resource 513 with which the target user has an interactive behavior can be obtained based on the historical behavior information of the target user 512 in the past period of time, and the target quantization code corresponding to any first historical resource 513 can be obtained by querying the first corresponding relationship 514, and then the first interest point is determined based on the interest point represented by the target quantization code, and the long-tail interest point 516 and the long-term interest point 515 of interest to the target user are obtained from the first interest point. Further, the resources under the long-tail interest point and the resources under the long-term interest point 517 can be obtained, and the recommendation model is requested to determine the target user's preference for the resources under the long-tail interest point of interest, and the target user's preference for the resources under the long-term interest point of interest, and then based on the preference, the target resource 518 is determined from the resources under the long-tail interest point and the resources under the long-term interest point 517, and then the target resource 518 can be recommended to the target user.

[0171] Therefore, resources under long-tail points of interest and long-term points of interest that the target user is interested in can be recommended to the target user, thereby solving the problem that the recommendation system is too concentrated in recommendation results due to the influence of system deviation, and cannot recommend resources under long-tail points of interest and long-term points of interest that the user is interested in, thereby meeting the user's demand for resources under long-tail points of interest and long-term points of interest, improving the accuracy and diversity of resource recommendations, optimizing the recommendation effect of resources, and improving user satisfaction. By using Transformer-Encoder as the user network, the user's characteristics can be more fully represented, and the user can be more fully understood, thereby improving the accuracy of the preference degree determined by the recommendation model. By performing quantitative encoding of resources during the model training process, real-time clustering of resources can be achieved, and then the corresponding relationship between resources and quantitative encoding can be obtained, and then based on the corresponding relationship, the long-term points of interest and long-tail points of interest that the target user is interested in can be quickly determined.

[0172] Combine the following Figure 6 , further illustrating the resource recommendation method provided in the embodiment of the present disclosure.

[0173] Figure 6 It is a flowchart of a resource recommendation method provided according to an embodiment of the present disclosure.

[0174] like Figure 6 As shown, the resource recommendation method includes:

[0175] Step 601: Acquire a first historical resource that has an interactive behavior with a target user.

[0176] Step 602: Acquire a first point of interest to which a first historical resource belongs.

[0177] Step 603: Acquire long-tail interest points that the target user is interested in from the first interest points.

[0178] The specific implementation process and principle of steps 601-603 may refer to the description of other embodiments and will not be repeated here.

[0179] Step 604: determine target search information from the first historical search information of the target user.

[0180] The first historical search information may include active search information of the target user in the past period of time, such as search terms entered by the target user in the search box of the Internet platform or search terms obtained by segmenting the search terms, etc. Active search information may accurately express the user's interests.

[0181] In some embodiments, search information that meets at least one of the following conditions in the first historical search information can be determined as target search information: the corresponding search frequency is higher than a frequency threshold, the corresponding search time is within a preset time period, and the corresponding number of search users is lower than a user number threshold. The search time corresponding to any search information is the time when the information is searched. The search user corresponding to any search information is the user who searches for the information.

[0182] Step 605: Determine the resources corresponding to the target search information.

[0183] Step 606: Recommend resources based on the resources under the long-tail interest points and the resources corresponding to the target search information.

[0184] In some embodiments, step 605 may be implemented by:

[0185] Acquire the information feature vector of the second history search information of the first history user, and acquire the resource feature vector of the second history resource with which the first history user has an interactive behavior;

[0186] For any information feature vector, determining the correlation between the information feature vector and the resource feature vector of the second historical resource;

[0187] Based on the correlation, a second corresponding relationship between the second historical search information and the second historical resource is obtained, and based on the second corresponding relationship, the resource corresponding to the target search information is determined.

[0188] The number of the first historical users may be one or at least two, which is not limited in the present disclosure.

[0189] The second historical search information is the search information of any first historical user in the past period of time.

[0190] Among them, the correlation between the information feature vector and the resource feature vector of the second historical resource can represent the correlation between the corresponding second historical search information and the second historical resource. In the disclosed embodiment, based on the correlation between the information feature vector and the resource feature vector, the second historical resource and the second historical search information with a higher correlation can be determined to obtain a second corresponding relationship between the second historical search information and the second historical resource. Furthermore, based on the second corresponding relationship, the second historical search information that is identical or similar to the target search information can be determined, and the second historical resource corresponding to the second historical search information can be determined as the resource corresponding to the target search information.

[0191] Thus, based on the correlation between the information feature vector of the second historical search information and the resource feature vector of the second historical resource, the second corresponding relationship between the second historical search information and the second historical resource can be quickly obtained. And since the second corresponding relationship is the corresponding relationship between the second historical search information and the second historical resource corresponding to the same first historical user, the second corresponding relationship can represent the association between the search information of the same first historical user and the resources of interest to him. Then, the resources corresponding to the target search information determined based on the second corresponding relationship are the resources of interest to the target user. Furthermore, by making resource recommendations based on the resources corresponding to the target search information, the target user can be recommended resources of interest, thereby improving the resource recommendation effect and user satisfaction.

[0192] In some embodiments, both the information feature vector and the resource feature vector are determined based on the second historical search information of the first historical user and the historical behavior information on the second historical resource.

[0193] That is, the information feature vector of the second historical search information of any first historical user can be determined based on the second historical search information of the first historical user and the historical behavior information of the first historical user on the second historical resource. The resource feature vector of any second historical resource that has an interactive behavior with the first historical user can be determined based on the second historical search information of the first historical user and the historical behavior information of the first historical user on the second historical resource.

[0194] Since both the information feature vector and the resource feature vector are determined based on the second historical search information of the first historical user and the historical behavior information on the second historical resource, not only can the information feature vector contain rich feature information of the second historical search information, and the resource feature vector contain rich feature information of the second historical resource, but the two feature vectors can also reflect the relationship between the second historical search information and the second historical resource. Then, the correlation between the information feature vector and the resource feature vector can more accurately reflect the correlation between the corresponding second historical search information and the second historical resource, so that the obtained second corresponding relationship can be more accurate, and the resource corresponding to the determined target search information can be more accurate.

[0195] In some embodiments, during the training process of a feature extraction network, such as when the training of the feature extraction network reaches a certain stage until the training is completed, the information feature vector of the second historical search information of the first historical user and the resource feature vector of the second historical resource with which the first historical user interacted can be obtained.

[0196] As a possible implementation manner, the feature extraction network may, for example, at least include a first feature extraction layer and a second feature extraction layer.

[0197] Among them, the input of the feature extraction network, that is, the input of the first feature extraction layer, may include the second historical search information of any first historical user and the historical behavior information of the first historical user on the second historical resource; the output of the first feature extraction layer, that is, the input of the second feature extraction layer, may include the information feature vector of the second historical search information and the resource feature vector of the second historical resource; the output of the second feature extraction layer, that is, the output of the feature extraction network, may include the user feature vector of the first historical user.

[0198] When training the feature extraction network, multiple training samples can be obtained, wherein each training sample includes the second historical search information of the same first historical user and the historical behavior information on the second historical resource, and the first historical user to which each training sample belongs is marked. Then, each training sample is input into the feature extraction network to obtain the user feature vector of the first historical user to which each training sample belongs. By substituting the user feature vector of the first historical user to which each training sample belongs into a preset loss function for loss calculation, a loss value can be obtained, and then the model parameters of the feature extraction network can be adjusted based on the loss value, so that the difference between the user feature vectors of the same first historical user is minimized and the difference between the user feature vectors of different first historical users is maximized, so that the feature extraction network learns how to determine the specific interest points and behavior patterns of each user, and learns to distinguish the differences between the interest points and behavior patterns of different users, and obtains a trained feature extraction network.

[0199] Among them, the information feature vector of any second historical search information and the resource feature vector of the second historical resource generated by the feature extraction network not only contain rich information respectively, such as the information feature vector contains the rich features of the second historical search information, and the resource feature vector contains the rich features of the second historical resource, but the two feature vectors can also reflect the relationship between the second historical search information and the second historical resource.

[0200] In some embodiments, two of the above feature extraction networks may be used to form a twin model, wherein each feature extraction network serves as a sub-network of the twin model.

[0201] Accordingly, any two training samples belonging to the same first historical user can be combined into a positive sample pair, and any two training samples belonging to different first historical users can be combined into a negative sample pair. In the process of training the twin model, the two training samples included in the positive sample pair can be respectively input into the two sub-networks of the twin model, or the two training samples included in the negative sample pair can be respectively input into the two sub-networks of the twin model, and the user feature vector of the first historical user to which the corresponding training sample belongs is output through each sub-network. Among them, each training sample does not need to be labeled as belonging to the first historical user. Then, the user feature vector of the first historical user to which each training sample belongs can be substituted into a preset loss function for loss calculation to obtain a loss value, and the model parameters of the twin model can be adjusted based on the loss value to minimize the difference between the user feature vectors of the same first historical user and maximize the difference between the user feature vectors of different first historical users, so that each sub-network of the twin model learns how to determine the specific interest points and behavior patterns of each user, and learns to distinguish the differences between the interest points and behavior patterns of different users, and obtains a trained twin model.

[0202] Among them, each sub-network of the twin model can generate an information feature vector of the second historical search information in the corresponding training sample, as well as a resource feature vector of the second historical resource. The information feature vector contains rich features of the second historical search information, and the resource feature vector contains rich features of the second historical resources. The two feature vectors can also reflect the relationship between the second historical search information and the second historical resources.

[0203] The resource recommendation method provided by the embodiment of the present disclosure can solve the problem that the recommendation system is too concentrated in the recommendation results due to the influence of system deviation, and cannot recommend resources under the long-tail interest points of interest to the user, thereby meeting the user's demand for resources under the long-tail interest points, improving the accuracy of resource recommendation, optimizing the resource recommendation effect, and improving user satisfaction. In addition, for a certain recommendation system, there may be users who have not interacted with the resources recommended by the recommendation system, or users who have interacted but less frequently. For such users, recommending resources based only on resources that have interacted with the user may result in poor recommendation effects. In the embodiment of the present disclosure, by determining the target search information from the first historical search information of the target user and then determining the resources corresponding to the target search information, resource recommendations are made based on the resources corresponding to the target search information. In this way, even if the number of first historical resources that have interactive behaviors with the target user is small, accurate resource recommendations can still be made to the target user based on the first historical search information of the target user, thereby improving the resource recommendation effect.

[0204] In order to implement the above embodiment, the present disclosure also provides a resource recommendation device.

[0205] Figure 7 It is a structural diagram of a resource recommendation device provided according to an embodiment of the present disclosure.

[0206] The resource recommendation device may be implemented in software and / or hardware, and may be configured in an electronic device, which may include but is not limited to a terminal, a server, and the like.

[0207] Among them, the resource recommendation device can be any Internet platform that requires individualized recommendations, such as a content distribution platform, an e-commerce platform, a social media platform, or can be configured in any of the above platforms, such as a recommendation system in any of the above platforms, and the present disclosure does not impose any restrictions on this.

[0208] like Figure 7 As shown, the resource recommendation device 700 includes:

[0209] A first acquisition module 701 is used to acquire a first historical resource that has an interactive behavior with a target user;

[0210] A second acquisition module 702 is used to acquire a first point of interest to which the first historical resource belongs;

[0211] A third acquisition module 703 is used to acquire the long-tail interest points that the target user is interested in from the first interest points;

[0212] The first recommendation module 704 is used to recommend resources based on the resources under the long-tail interest points.

[0213] As a possible implementation of the embodiment of the present disclosure, the second acquisition module 702 includes:

[0214] A first acquisition unit, configured to acquire a target quantization code corresponding to the first historical resource based on a first corresponding relationship between a first preset resource and a quantization code, wherein the target quantization code represents a point of interest to which the first historical resource belongs;

[0215] A first determining unit is configured to determine the first interest point based on the interest point represented by the target quantization code.

[0216] As another possible implementation of the embodiment of the present disclosure, the number of the first preset resources is at least two, and the resource recommendation device 700 further includes:

[0217] A first clustering module, configured to cluster the first preset resources based on similarities between features of the first preset resources to obtain a first cluster to which the first preset resources belong, wherein the first cluster represents a type of a point of interest to which the first preset resource belongs;

[0218] A first determining module, configured to determine a first distance between the first preset resource and a cluster center of the first cluster to which the first preset resource belongs;

[0219] A second clustering module, configured to cluster the first distance to obtain a second cluster to which the first distance belongs, wherein the second cluster represents a subtype to which the first preset resource belongs under the type of the corresponding point of interest;

[0220] A second determining module, configured to obtain a first code of a cluster center of the first cluster and a second code of a cluster center of the second cluster, and determine a quantized code representing a point of interest to which the first preset resource belongs based on the first code and the second code;

[0221] The third determination module is used to determine the first corresponding relationship based on the quantization code corresponding to the first preset resource.

[0222] As another possible implementation of the embodiment of the present disclosure, the resource recommendation device 700 further includes:

[0223] a fourth determining module, configured to determine, from the first cluster, a third cluster to which the second preset resource belongs based on a similarity between features of the second preset resource and the cluster center of the first cluster, wherein the third cluster represents a type of the point of interest to which the second preset resource belongs;

[0224] a fifth determining module, configured to determine a second distance between the second preset resource and the cluster center of the third cluster to which it belongs, and determine, from the second cluster, a fourth cluster to which the second distance belongs based on a similarity between the second distance and the cluster center of the second cluster, wherein the fourth cluster represents a subtype to which the second preset resource belongs under the type of the point of interest to which it belongs;

[0225] a sixth determining module, configured to obtain a third code of the cluster center of the third cluster and a fourth code of the cluster center of the fourth cluster, and determine a quantized code representing a point of interest to which the second preset resource belongs based on the third code and the fourth code;

[0226] The first updating module is used to update the first corresponding relationship based on the quantization code corresponding to the second preset resource.

[0227] As another possible implementation of the embodiment of the present disclosure, the resource recommendation device 700 further includes:

[0228] A second updating module, configured to update the cluster center of the third cluster based on the second preset resource and the first preset resource included in the third cluster;

[0229] The third updating module is configured to update the cluster center of the fourth cluster based on the second distance and the first distance included in the fourth cluster.

[0230] As another possible implementation of the embodiment of the present disclosure, the third acquisition module 703 includes:

[0231] A second acquisition unit is used to acquire the number of clicks on resources under at least two preset points of interest;

[0232] A second determining unit is configured to determine a second point of interest from the preset points of interest based on the number of clicks, wherein the number of clicks on resources under the second point of interest is lower than the number of clicks on resources under other points of interest among the preset points of interest except the second point of interest;

[0233] The third determining unit is configured to determine the interest points among the first interest points that belong to the second interest points as the long-tail interest points.

[0234] As another possible implementation of the embodiment of the present disclosure, the first recommendation module 704 includes:

[0235] A fourth determining unit, configured to determine the target user's preference for the resources under the long-tail interest point through a recommendation model;

[0236] A recommendation unit is used to determine a target resource from the resources under the long-tail interest point based on the preference degree, and recommend the target resource to the target user.

[0237] As another possible implementation manner of the embodiment of the present disclosure, the fourth determining unit is configured to:

[0238] Acquire the first characteristic information of the target user and the second characteristic information of the resources under the long-tail interest point;

[0239] Inputting the first feature information into the user network used to generate a user feature vector in the recommendation model to obtain the user feature vector of the target user;

[0240] Inputting the second feature information into a resource network used to generate a resource feature vector in the recommendation model to obtain a resource feature vector of the resource under the long-tail interest point;

[0241] The preference degree is determined based on the obtained user feature vector and the resource feature vector.

[0242] As another possible implementation of the embodiment of the present disclosure, the resource recommendation device 700 further includes:

[0243] A fourth acquisition module, configured to acquire long-term interest points of interest to the target user from the first interest points;

[0244] The second recommendation module is used to recommend resources based on the resources under the long-term interest points.

[0245] As another possible implementation of the embodiment of the present disclosure, the fourth acquisition module includes:

[0246] The fifth determining unit is used to determine the quantity of the first historical resources under the first point of interest, and determine the first point of interest as the long-term point of interest when the quantity is higher than a quantity threshold.

[0247] As another possible implementation of the embodiment of the present disclosure, the resource recommendation device 700 further includes:

[0248] A seventh determination module, configured to determine target search information from the first historical search information of the target user;

[0249] An eighth determination module, used to determine the resource corresponding to the target search information;

[0250] The third recommendation module is used to recommend resources based on the resources corresponding to the target search information.

[0251] As another possible implementation of the embodiment of the present disclosure, the eighth determining module includes:

[0252] A third acquisition unit is used to acquire the information feature vector of the second historical search information of the first historical user, and acquire the resource feature vector of the second historical resource with which the first historical user has an interactive behavior;

[0253] a sixth determining unit, configured to determine a correlation between the information feature vector and a resource feature vector of the second historical resource;

[0254] A seventh determining unit is configured to obtain a second corresponding relationship between the second historical search information and the second historical resource based on the correlation, and determine a resource corresponding to the target search information based on the second corresponding relationship.

[0255] As another possible implementation manner of the embodiment of the present disclosure, both the information feature vector and the resource feature vector are determined according to the second historical search information of the first historical user and the historical behavior information on the second historical resource.

[0256] As another possible implementation of the embodiment of the present disclosure, the first recommendation module 704 is used to:

[0257] When it is determined that the resources recommended to the second historical user do not include resources under the long-tail interest point that the second historical user is interested in, recommending resources based on the resources under the long-tail interest point;

[0258] Among them, the long-tail interest points of the second historical user are obtained based on the historical behavior information of the second historical user on the third historical resource, and the third historical resource is a resource obtained from a resource queue and recommended to the second historical user, and the resource queue includes resources randomly obtained from a resource library.

[0259] It should be noted that the above explanation of the resource recommendation method is also applicable to the resource recommendation device of this embodiment and will not be repeated here.

[0260] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0261] Figure 8A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0262] like Figure 8 As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 802 or a computer program loaded from a storage unit 805 to a RAM (Random Access Memory) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An I / O (Input / Output) interface 805 is also connected to the bus 804.

[0263] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 805, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0264] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 805. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the resource recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the method resource recommendation method in any other appropriate manner (eg, by means of firmware).

[0265] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0266] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0267] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0268] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0269] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0270] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0271] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0272] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0273] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A resource recommendation method, comprising: Acquire the first historical resource that has interactive behavior with the target user; Acquire a first point of interest to which the first historical resource belongs; Acquire the long-tail interest points that the target user is interested in from the first interest points; Resource recommendations are made based on the resources under the long-tail interest points.

2. The method according to claim 1, wherein: The acquiring the first point of interest to which the first historical resource belongs includes: Based on a first correspondence between a first preset resource and a quantization code, obtaining a target quantization code corresponding to the first historical resource, wherein the target quantization code represents a point of interest to which the first historical resource belongs; The first interest point is determined based on the interest point represented by the target quantization code.

3. The method according to claim 2, wherein: The number of the first preset resources is at least two, and the method further includes: Based on the similarity between the features of the first preset resources, clustering the first preset resources to obtain a first cluster to which the first preset resources belong, wherein the first cluster represents the type of the point of interest to which the first preset resource belongs; Determine a first distance between the first preset resource and a cluster center of the first cluster to which it belongs, and cluster the first distance to obtain a second cluster to which the first distance belongs, wherein the second cluster represents a subtype to which the first preset resource belongs under the type of the point of interest to which it belongs; Acquire a first code of a cluster center of the first cluster and a second code of a cluster center of the second cluster, and determine a quantized code representing a point of interest to which the first preset resource belongs based on the first code and the second code; The first corresponding relationship is determined based on the quantization code corresponding to the first preset resource.

4. The method according to claim 3, wherein: The method further comprises: Based on the similarity between the features of the second preset resource and the cluster center of the first cluster, determining from the first cluster a third cluster to which the second preset resource belongs, wherein the third cluster represents the type of the point of interest to which the second preset resource belongs; Determine a second distance between the second preset resource and the cluster center of the third cluster to which it belongs, and determine, from the second cluster based on the similarity between the second distance and the cluster center of the second cluster, a fourth cluster to which the second distance belongs, wherein the fourth cluster represents a subtype to which the second preset resource belongs under the type of the point of interest to which it belongs; Acquire a third code of the cluster center of the third cluster and a fourth code of the cluster center of the fourth cluster, and determine a quantized code representing the point of interest to which the second preset resource belongs based on the third code and the fourth code; The first corresponding relationship is updated based on the quantization code corresponding to the second preset resource.

5. The method according to claim 4, wherein: The method further comprises: Based on the second preset resource and the first preset resource included in the third cluster, updating the cluster center of the third cluster; The cluster center of the fourth cluster is updated based on the second distance and the first distance included in the fourth cluster.

6. The method according to claim 1, wherein: The acquiring the long-tail interest points that the target user is interested in from the first interest points includes: Obtain the number of clicks on resources under at least two preset points of interest; Based on the click volume, determining a second point of interest from the preset points of interest, wherein the click volume of resources under the second point of interest is lower than the click volume of resources under other points of interest among the preset points of interest except the second point of interest; The interest points among the first interest points that belong to the second interest points are determined as the long-tail interest points.

7. The method according to any one of claims 1 to 6, wherein: The recommending resources based on the resources under the long-tail interest points includes: Determining the target user's preference for the resources under the long-tail interest point through a recommendation model; Based on the preference degree, a target resource is determined from the resources under the long-tail interest point, and the target resource is recommended to the target user.

8. The method according to claim 7, wherein: Determining the target user's preference for the resources under the long-tail interest point through the recommendation model includes: Acquire the first characteristic information of the target user and the second characteristic information of the resources under the long-tail interest point; Inputting the first feature information into the user network used to generate a user feature vector in the recommendation model to obtain the user feature vector of the target user; Inputting the second feature information into a resource network used to generate a resource feature vector in the recommendation model to obtain a resource feature vector of the resource under the long-tail interest point; The preference degree is determined based on the obtained user feature vector and the resource feature vector.

9. The method according to any one of claims 1 to 6, wherein: The method further comprises: Acquire the long-term interest points that the target user is interested in from the first interest points; Resource recommendations are made based on the resources under the long-term interest points.

10. The method according to claim 9, wherein: The acquiring the long-term interest point that the target user is interested in from the first interest point includes: Determining the number of first historical resources under the first point of interest; In a case where the number is higher than a number threshold, the first point of interest is determined as the long-term point of interest.

11. The method according to any one of claims 1 to 6, wherein: The method further comprises: Determining target search information from the first historical search information of the target user; Determining resources corresponding to the target search information; Recommend resources based on the resources corresponding to the target search information.

12. The method according to claim 11, wherein: The determining of the resource corresponding to the target search information includes: Acquire an information feature vector of second history search information of a first history user, and acquire a resource feature vector of a second history resource with which the first history user has an interactive behavior; Determining a correlation between the information feature vector and a resource feature vector of the second historical resource; Based on the correlation, a second corresponding relationship between the second historical search information and the second historical resource is obtained, and based on the second corresponding relationship, a resource corresponding to the target search information is determined.

13. The method according to claim 12, wherein: The information feature vector and the resource feature vector are both determined based on the second historical search information of the first historical user and the historical behavior information on the second historical resource.

14. The method according to any one of claims 1 to 6, wherein: The recommending resources based on the resources under the long-tail interest points includes: When it is determined that the resources recommended to the second historical user do not include resources under the long-tail interest point that the second historical user is interested in, recommending resources based on the resources under the long-tail interest point; Among them, the long-tail interest points of the second historical user are obtained based on the historical behavior information of the second historical user on the third historical resource, and the third historical resource is a resource obtained from a resource queue and recommended to the second historical user, and the resource queue includes resources randomly obtained from a resource library.

15. A resource recommendation device, comprising: A first acquisition module, used to acquire a first historical resource having an interactive behavior with a target user; A second acquisition module, used to acquire a first point of interest to which the first historical resource belongs; A third acquisition module is used to acquire the long-tail interest points that the target user is interested in from the first interest points; The first recommendation module is used to recommend resources based on the resources under the long-tail interest points.

16. The device according to claim 15, wherein: The second acquisition module includes: A first acquisition unit, configured to acquire a target quantization code corresponding to the first historical resource based on a first corresponding relationship between a first preset resource and a quantization code, wherein the target quantization code represents a point of interest to which the first historical resource belongs; A first determining unit is configured to determine the first interest point based on the interest point represented by the target quantization code.

17. The device according to claim 16, wherein: The number of the first preset resources is at least two, and the device further includes: A first clustering module, configured to cluster the first preset resources based on similarities between features of the first preset resources to obtain a first cluster to which the first preset resources belong, wherein the first cluster represents a type of a point of interest to which the first preset resource belongs; A first determining module, configured to determine a first distance between the first preset resource and a cluster center of the first cluster to which the first preset resource belongs; A second clustering module, configured to cluster the first distance to obtain a second cluster to which the first distance belongs, wherein the second cluster represents a subtype to which the first preset resource belongs under the type of the corresponding point of interest; A second determining module, configured to obtain a first code of a cluster center of the first cluster and a second code of a cluster center of the second cluster, and determine a quantized code representing a point of interest to which the first preset resource belongs based on the first code and the second code; The third determination module is used to determine the first corresponding relationship based on the quantization code corresponding to the first preset resource.

18. The device according to claim 17, wherein: The device also includes: a fourth determining module, configured to determine, from the first cluster, a third cluster to which the second preset resource belongs based on a similarity between features of the second preset resource and the cluster center of the first cluster, wherein the third cluster represents a type of the point of interest to which the second preset resource belongs; a fifth determining module, configured to determine a second distance between the second preset resource and the cluster center of the third cluster to which it belongs, and determine, from the second cluster, a fourth cluster to which the second distance belongs based on a similarity between the second distance and the cluster center of the second cluster, wherein the fourth cluster represents a subtype to which the second preset resource belongs under the type of the point of interest to which it belongs; a sixth determining module, configured to obtain a third code of the cluster center of the third cluster and a fourth code of the cluster center of the fourth cluster, and determine a quantized code representing a point of interest to which the second preset resource belongs based on the third code and the fourth code; The first updating module is used to update the first corresponding relationship based on the quantization code corresponding to the second preset resource.

19. The device according to claim 18, wherein: The device also includes: A second updating module, configured to update the cluster center of the third cluster based on the second preset resource and the first preset resource included in the third cluster; The third updating module is configured to update the cluster center of the fourth cluster based on the second distance and the first distance included in the fourth cluster.

20. The device according to claim 15, wherein: The third acquisition module includes: A second acquisition unit is used to acquire the number of clicks on resources under at least two preset points of interest; A second determining unit is configured to determine a second point of interest from the preset points of interest based on the number of clicks, wherein the number of clicks on resources under the second point of interest is lower than the number of clicks on resources under other points of interest among the preset points of interest except the second point of interest; The third determining unit is configured to determine the interest points among the first interest points that belong to the second interest points as the long-tail interest points.

21. The device according to any one of claims 15 to 20, wherein: The first recommendation module includes: A fourth determining unit, configured to determine the target user's preference for the resources under the long-tail interest point through a recommendation model; A recommendation unit is used to determine a target resource from the resources under the long-tail interest point based on the preference degree, and recommend the target resource to the target user.

22. The device according to claim 21, wherein The fourth determining unit is configured to: Acquire the first characteristic information of the target user and the second characteristic information of the resources under the long-tail interest point; Inputting the first feature information into the user network used to generate a user feature vector in the recommendation model to obtain the user feature vector of the target user; Inputting the second feature information into a resource network used to generate a resource feature vector in the recommendation model to obtain a resource feature vector of the resource under the long-tail interest point; The preference degree is determined based on the obtained user feature vector and the resource feature vector.

23. The device according to any one of claims 15 to 20, wherein: The device also includes: A fourth acquisition module, configured to acquire long-term interest points of interest to the target user from the first interest points; The second recommendation module is used to recommend resources based on the resources under the long-term interest points.

24. The device according to claim 23, wherein: The fourth acquisition module includes: The fifth determining unit is used to determine the quantity of the first historical resources under the first point of interest, and determine the first point of interest as the long-term point of interest when the quantity is higher than a quantity threshold.

25. The device according to any one of claims 15 to 20, wherein: The device also includes: A seventh determination module, configured to determine target search information from the first historical search information of the target user; An eighth determination module, used to determine the resource corresponding to the target search information; The third recommendation module is used to recommend resources based on the resources corresponding to the target search information.

26. The device according to claim 25, wherein The eighth determination module comprises: A third acquisition unit is used to acquire the information feature vector of the second historical search information of the first historical user, and acquire the resource feature vector of the second historical resource with which the first historical user has an interactive behavior; a sixth determining unit, configured to determine a correlation between the information feature vector and a resource feature vector of the second historical resource; A seventh determining unit is configured to obtain a second corresponding relationship between the second historical search information and the second historical resource based on the correlation, and determine a resource corresponding to the target search information based on the second corresponding relationship.

27. The device according to claim 26, wherein: The information feature vector and the resource feature vector are both determined based on the second historical search information of the first historical user and the historical behavior information on the second historical resource.

28. The device according to any one of claims 15 to 20, wherein: The first recommendation module is used to: When it is determined that the resources recommended to the second historical user do not include resources under the long-tail interest point that the second historical user is interested in, recommending resources based on the resources under the long-tail interest point; Among them, the long-tail interest points of the second historical user are obtained based on the historical behavior information of the second historical user on the third historical resource, and the third historical resource is a resource obtained from a resource queue and recommended to the second historical user, and the resource queue includes resources randomly obtained from a resource library.

29. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 14.

30. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 14.

31. A computer program product, comprising a computer program, which, when executed by a processor, implements the method of any one of claims 1 to 14.

Images