Resource recommendation model training method, resource recommendation method and device

Abstract

The present disclosure provides a method for training a resource recommendation model, which can be applied to the field of artificial intelligence technology. The training method of the resource recommendation model includes: obtaining a resource database, wherein the resource database includes multiple clusters obtained by pre-clustering, and each cluster includes at least one resource; determining a target cluster from multiple clusters; obtaining first description information of users associated with resources in the target cluster, and second description information of resources in the target cluster, wherein the first description information has label information, and the label information represents the probability of users selecting resources in the target cluster; and using the first description information, label information, and second description information to train the resource recommendation model to be trained to generate a trained resource recommendation model. The present disclosure also provides a training device, resource recommendation method, resource recommendation device, equipment, storage medium, and program product for a resource recommendation model.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically to a resource recommendation model training method, resource recommendation method, device, equipment, medium, and program product. Background Art

[0002] The emergence and widespread use of the internet has brought users a vast amount of data and information, satisfying their demand for data and information in the information age. However, the explosive growth of data volumes brought about by the rapid development of the internet has made it difficult for users to obtain effective information from this vast amount of information, resulting in a decrease in their efficiency in using information and leading to the problem of information overload.

[0003] The resource recommendation method is a method to solve the problem that users cannot quickly find information that meets their interests and needs due to information overload.

[0004] In the process of realizing the concept of the present disclosure, the inventors discovered that the resource recommendation method in the related art has a technical problem of low accuracy. Summary of the Invention

[0005] In view of the above problems, the present disclosure provides a resource recommendation model training method, resource recommendation method, device, equipment, medium and program product.

[0006] According to a first aspect of the present disclosure, a method for training a resource recommendation model is provided, comprising:

[0007] Acquire a resource database, wherein the resource database includes a plurality of clusters obtained through pre-clustering, and each cluster includes at least one resource;

[0008] Determine a target cluster from the plurality of clusters mentioned above;

[0009] Obtaining first description information of a user associated with a resource in the target cluster, and second description information of the resource in the target cluster, wherein the first description information represents the user's preference for the resource, and the first description information has tag information, and the tag information represents the user's selection probability for the resource in the target cluster; and

[0010] The resource recommendation model to be trained is trained using the first description information, the label information, and the second description information to generate a trained resource recommendation model.

[0011] According to an embodiment of the present disclosure, the resource database is generated by the following operations:

[0012] Generating a statistical data matrix based on quality information of each resource in the initial resource database, wherein the quality information is obtained based on log data generated by user operations on the resource;

[0013] Perform data processing on the above statistical data matrix to generate a standardized matrix;

[0014] Performing fuzzy similarity calculation on the data in the above-mentioned standardized matrix to generate a fuzzy similarity matrix;

[0015] Based on the above fuzzy similarity matrix, a fuzzy equivalence matrix is ​​generated;

[0016] Based on the fuzzy equivalence matrix, the resources in the initial resource database are clustered to generate the resource database.

[0017] According to an embodiment of the present disclosure, clustering the resources in the initial resource database based on the fuzzy equivalence matrix to generate the resource database includes:

[0018] Obtain multiple thresholds to be classified from the above fuzzy equivalence matrix;

[0019] The resources in the initial resource database are clustered according to each of the thresholds to be classified to generate the resource database.

[0020] According to an embodiment of the present disclosure, generating a fuzzy equivalence matrix based on the fuzzy similarity matrix includes:

[0021] Iteratively do the following:

[0022] Determine whether the fuzzy similarity matrix meets the iteration stop condition;

[0023] When the fuzzy similarity matrix does not satisfy the iteration stopping condition, performing a square operation on the fuzzy similarity matrix to generate a first fuzzy similarity matrix; and

[0024] Using the first fuzzy similarity matrix as the fuzzy similarity matrix; and

[0025] When the fuzzy similarity matrix satisfies the iteration stopping condition, the fuzzy similarity matrix is ​​used as the fuzzy equivalent matrix.

[0026] According to an embodiment of the present disclosure, a row of data in the above-mentioned standardized matrix represents one of the above-mentioned resources;

[0027] The fuzzy similarity calculation is performed on the data in the above-mentioned standardized matrix to generate a fuzzy similarity matrix, which includes:

[0028] Calculate the similarity between any two rows of data in the standardized data matrix until every row of data in the standardized data matrix is ​​traversed to generate a similarity result;

[0029] The fuzzy similarity matrix is ​​generated according to the similarity results.

[0030] According to an embodiment of the present disclosure, the step of processing the statistical data matrix to generate a standardized matrix includes:

[0031] The statistical data matrix is ​​normalized to generate a first matrix;

[0032] The first matrix is ​​processed by translation range change to generate the standardized matrix.

[0033] According to an embodiment of the present disclosure, the step of determining a target cluster from the plurality of clusters includes:

[0034] Obtain classification description data for each of the above clusters;

[0035] Calculate statistics based on each of the above classification description data to generate multiple statistical results;

[0036] The cluster corresponding to the maximum value of the statistical result is determined as the target cluster.

[0037] According to an embodiment of the present disclosure, the method of training the resource recommendation model to be trained using the first description information and the second description information to generate a trained resource recommendation model includes:

[0038] Inputting the first description information and the second description information into the resource recommendation model to be trained, and outputting a prediction result, wherein the prediction result represents the predicted click probability of the target object to be recommended for the resource corresponding to the second description information;

[0039] Input the above prediction results and the above label information into the loss function and output the loss result;

[0040] The network parameters of the resource recommendation model to be trained are iteratively adjusted according to the loss result to generate the trained resource recommendation model.

[0041] According to an embodiment of the present disclosure, in the case where the above-mentioned resource is a video resource, the above-mentioned quality information includes the number of plays, the number of likes, the number of comments, and the number of reposts of the above-mentioned video resource.

[0042] A second aspect of the present disclosure provides a resource recommendation method, comprising:

[0043] Inputting description information of the object to be recommended into a resource recommendation model and outputting a prediction result array, wherein the prediction result array includes a plurality of prediction probability values, each of which represents a click probability of the object to be recommended for at least one resource, and the resource recommendation model is trained using the resource recommendation model training method;

[0044] Based on the above prediction result array, resource recommendations are made for the above objects to be recommended.

[0045] A third aspect of the present disclosure provides a training device for a resource recommendation model, comprising:

[0046] A first acquisition module is configured to acquire a resource database, wherein the resource database includes a plurality of clusters obtained through pre-clustering, and each cluster includes at least one resource;

[0047] A first determining module is used to determine a target cluster from the plurality of clusters mentioned above;

[0048] a second acquisition module, configured to acquire first description information of a user associated with a resource in the target cluster, and second description information of the resource in the target cluster, wherein the first description information represents the user's preference for the resource, and the first description information has tag information, and the tag information represents the user's selection probability for the resource in the target cluster; and

[0049] The training module is used to train the resource recommendation model to be trained using the first description information, the label information and the second description information to generate a trained resource recommendation model.

[0050] A fourth aspect of the present disclosure provides a resource recommendation device, comprising:

[0051] An input module, configured to input description information of a to-be-recommended object into a resource recommendation model and output a prediction result array, wherein the prediction result array includes a plurality of prediction probability values, each of which represents a click probability of the to-be-recommended object for at least one resource, the resource recommendation model being trained using the resource recommendation model training method;

[0052] The recommendation module is used to recommend resources to the above-mentioned objects to be recommended based on the above-mentioned prediction result array.

[0053] The fifth aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned resource recommendation model training method and resource recommendation method.

[0054] The sixth aspect of the present disclosure further provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above-mentioned resource recommendation model training method and resource recommendation method.

[0055] The seventh aspect of the present disclosure further provides a computer program product, including a computer program, which, when executed by a processor, implements the training method and resource recommendation method of the resource recommendation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0057] Figure 1 A diagram schematically illustrates an application scenario of a resource recommendation model training method, a resource recommendation model training device, a resource recommendation method, and a resource recommendation device according to an embodiment of the present disclosure;

[0058] Figure 2 The flowchart of the training method of the resource recommendation model according to the embodiment of the present disclosure is schematically shown;

[0059] Figure 3 Schematically shows a flow chart of generating a resource database according to an embodiment of the present disclosure;

[0060] Figure 4 A flowchart of performing fuzzy similarity calculation on data in a standardized matrix to generate a fuzzy similarity matrix according to an embodiment of the present disclosure is schematically shown;

[0061] Figure 5 A flowchart of generating a fuzzy equivalence matrix based on a fuzzy similarity matrix according to an embodiment of the present disclosure is schematically shown;

[0062] Figure 6 A flowchart of clustering resources in an initial resource database based on a fuzzy equivalence matrix to generate a resource database according to an embodiment of the present disclosure is schematically shown;

[0063] Figure 7 A flowchart of determining a target cluster from multiple clusters according to an embodiment of the present disclosure is schematically shown;

[0064] Figure 8 Schematically shows a flow chart of processing a statistical data matrix to generate a standardized matrix according to an embodiment of the present disclosure;

[0065] Figure 9 A flowchart schematically illustrates how to train a resource recommendation model to be trained using first description information and second description information to generate a trained resource recommendation model according to an embodiment of the present disclosure;

[0066] Figure 10 The following schematically shows a flow chart of a resource recommendation method according to an embodiment of the present disclosure;

[0067] Figure 11 The following schematically shows a structural block diagram of a training device for a resource recommendation model according to an embodiment of the present disclosure;

[0068] Figure 12 Schematically shows a structural block diagram of a resource recommendation device according to an embodiment of the present disclosure; and

[0069] Figure 13 The block diagram schematically shows an electronic device suitable for implementing a resource recommendation model training method and a resource recommendation method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0070] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0071] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0072] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0073] When expressions such as "at least one of A, B and C, etc." are used, they should generally be interpreted in accordance with the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0074] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0075] The emergence and popularization of the Internet has brought a large amount of data and information to users, satisfying their demand for data and information in the network information age. However, with the rapid development of the Internet, the amount of data has exploded, making it difficult for users to obtain effective information from the large amount of information, and their efficiency in using information has decreased. This is the so-called information overload problem.

[0076] "Internet + Education" is a new form of education that combines Internet technology with the field of education with the continuous development of science and technology today. It integrates emerging technologies such as VR (Virtual Reality), big data, and artificial intelligence, enabling education to move from assembly line to personalized.

[0077] Similarly, the "Internet + Education" sector faces information overload. Faced with a vast amount of learning resources, users often struggle to accurately locate the resources they need. Therefore, a highly promising solution to this problem is recommendation systems. Recommendation systems are software tools and technologies that suggest resources of potential interest to target users. They recommend information and products based on their information needs and interests.

[0078] The present disclosure provides a method for training a resource recommendation model, which can be applied to the field of artificial intelligence technology. The training method of the resource recommendation model includes: obtaining a resource database, wherein the resource database includes multiple clusters obtained by pre-clustering, and each cluster includes at least one resource; determining a target cluster from multiple clusters; obtaining first description information of users associated with resources in the target cluster, and second description information of resources in the target cluster, wherein the first description information has label information, and the label information represents the probability of users selecting resources in the target cluster; and using the first description information, label information, and second description information to train the resource recommendation model to be trained to generate a trained resource recommendation model. The present disclosure also provides a training device, resource recommendation method, resource recommendation device, equipment, storage medium, and program product for a resource recommendation model.

[0079] Figure 1 The application scenario diagram of the resource recommendation model training method, resource recommendation model training device, resource recommendation method, and resource recommendation device according to the embodiments of the present disclosure is schematically shown.

[0080] like Figure 1As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0081] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0082] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0083] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process received data such as user requests, and feed back processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0084] It should be noted that the training method of the resource recommendation model and the resource recommendation method provided in the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the training device of the resource recommendation model and the resource recommendation device provided in the embodiments of the present disclosure can generally be set in the server 105. The training method of the resource recommendation model and the resource recommendation method provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the training device of the resource recommendation model and the resource recommendation device provided in the embodiments of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0085] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0086] The following will be based on Figure 1 The scene described by Figures 2 to 9The training method of the resource recommendation model of the embodiment of the present disclosure is described in detail.

[0087] Figure 2 The flowchart of the resource recommendation model training method according to the embodiment of the present disclosure is schematically shown.

[0088] like Figure 2 As shown, the resource recommendation model training method of this embodiment includes operations S201 to S204.

[0089] In operation S201 , a resource database is acquired, wherein the resource database includes a plurality of clusters obtained through pre-clustering, and each cluster includes at least one resource.

[0090] According to an embodiment of the present disclosure, resources may include, for example, video resources, voice resources, image resources, text resources, etc.

[0091] According to an embodiment of the present disclosure, multiple resources may be clustered in an offline clustering manner to form multiple clusters, each cluster including at least one resource.

[0092] In operation S202 , a target cluster is determined from a plurality of clusters.

[0093] In operation S203, first description information of the user associated with the resources in the target cluster and second description information of the resources in the target cluster are obtained, wherein the first description information represents the user's preference for the resources, and the first description information has label information, and the label information represents the user's selection probability for the resources in the target cluster.

[0094] According to an embodiment of the present disclosure, the second description information may characterize the attributes of the resources in the target cluster, for example, the size of the occupied storage space, type, duration, target audience, whether it is charged, etc.

[0095] According to an embodiment of the present disclosure, during the training stage of the resource recommendation model, the first description information of the user who has selected resources in the target cluster within a preset historical time period can be obtained. Since the user has a certain association relationship with the resources in the target cluster, the probability of the user selecting the resources in the target cluster can be calculated, and the probability value can be used as the label information of the first description information.

[0096] In operation S204 , the resource recommendation model to be trained is trained using the first description information, the tag information, and the second description information to generate a trained resource recommendation model.

[0097] According to an embodiment of the present disclosure, a supervised or semi-supervised training method can be adopted to train the resource recommendation model to be trained using the first description information, label information and second description information to generate a trained resource recommendation model.

[0098] In an embodiment of the present disclosure, by selecting a target cluster from multiple pre-clustered clusters, and then using the second description information of the resources in the target cluster and the first description information of the user to train a resource recommendation model, the resource recommendation accuracy of the trained resource recommendation model can be improved.

[0099] Figure 3 The flowchart of generating a resource database according to an embodiment of the present disclosure is schematically shown.

[0100] like Figure 3 As shown, the method of this embodiment includes operations S301 to S305.

[0101] In operation S301 , a statistical data matrix is ​​generated according to quality information of each resource in an initial resource database, wherein the quality information is obtained according to log data generated by a user's operation on a resource.

[0102] According to an embodiment of the present disclosure, in the process of recommending resources, it is necessary to recommend high-quality resources that meet the user's expectations based on the user's behavior while ensuring the content quality of the recommended resources.

[0103] For the convenience of discussion, the training method of the resource recommendation model provided by the embodiment of the present disclosure is explained below using recommended video resources. However, it should be noted that the training method of the resource recommendation model provided by the embodiment of the present disclosure is not only applicable to training resource recommendation models for recommending video resources, but can also be applied to training resource recommendation models for recommending image resources, voice resources or text resources, etc.

[0104] For video resources, data such as viewing time, likes, comments, and reposts are the main sources of content traffic. Therefore, when recommending video resources, we can analyze user behavior log data to obtain statistics on the number of plays, likes, comments, and reposts for each video resource in the initial resource database, and use the number of plays, likes, comments, and reposts as quality information for the video resource.

[0105] According to an embodiment of the present disclosure, the statistical data matrix can be expressed by the following formula (1).

[0106]

[0107] Among them, VQ represents the statistical data matrix, V play Indicates the number of plays, V like Indicates the number of likes, V repost Represents the number of forwarding, V comment Indicates the number of comments.

[0108] In operation S302 , data processing is performed on the statistical data matrix to generate a standardized matrix.

[0109] According to an embodiment of the present disclosure, since the number of plays, likes, comments, and reposts in the statistical data matrix are usually not data of the same data dimension, for example, the number of plays may be thousands, the number of reposts may be dozens, etc., and such data of different data dimensions may cause difficulties in the subsequent data processing process, after generating the statistical data matrix, the statistical data matrix can be first processed, and the data in the statistical data matrix can be standardized to generate a standardized matrix.

[0110] In operation S303 , fuzzy similarity calculation is performed on the data in the normalized matrix to generate a fuzzy similarity matrix.

[0111] In operation S304 , a fuzzy equivalence matrix is ​​generated based on the fuzzy similarity matrix.

[0112] In operation S305 , resources in the initial resource database are clustered based on the fuzzy equivalence matrix to generate a resource database.

[0113] Figure 4 The flowchart of performing fuzzy similarity calculation on data in a standardized matrix and generating a fuzzy similarity matrix according to an embodiment of the present disclosure is schematically shown.

[0114] According to an embodiment of the present disclosure, a row of data in the normalized matrix represents one resource.

[0115] like Figure 4 As shown, the method of this embodiment includes operations S401 to S402.

[0116] In operation S401 , similarity calculation is performed between any two rows of data in the standardized data matrix until every row of data in the standardized data matrix is ​​traversed to generate a similarity result.

[0117] In operation S402 , a fuzzy similarity matrix is ​​generated according to the similarity results.

[0118] According to an embodiment of the present disclosure, since a row of data in a standardized matrix represents a resource, similarity calculation can be performed between any two rows of data in the standardized data matrix. For example, the similarity calculation can be performed between the first row and the first row first to generate a first similarity result, and the first similarity result is used as the data at the position (1, 1) of the fuzzy similarity matrix. Then, the similarity calculation can be performed between the first row and the second row to generate a second similarity result, and the second similarity result is used as the data at the position (1, 2) of the fuzzy similarity matrix, until each row of data in the standardized data matrix is ​​traversed.

[0119] According to an embodiment of the present disclosure, the fuzzy similarity matrix can be expressed as S(vq x , vq y ), where the fuzzy similarity matrix S consists of elements vq xy The elements of this matrix represent any two video resources vq x and vq y The similarity distance s between xy , the specific calculation of similarity distance is shown in formula (2).

[0120] s xy =1-d(vq x , vq y ); (2)

[0121] For everything xy All meet: |s xy |≤1,|s xx |=1,s xy =s yx , where 0≤s xy ≤1,s xy The closer the value is to 1, the higher the vq x and vq y The closer the relationship between them, the higher the similarity; the diagonal element value of the fuzzy similarity matrix is ​​1, which means that the video resource has the greatest similarity with itself.

[0122] where d(vq x , vq y ) indicates the video resource vq x Quality information and video resources vq y The similarity distance between the quality information, in the embodiment of the present disclosure, the similarity distance can be calculated by Euclidean distance, and the calculation method is shown in formula (3);

[0123] Where (1≤x, y≤n); (3)

[0124] Figure 5 The flowchart of generating a fuzzy equivalence matrix based on a fuzzy similarity matrix according to an embodiment of the present disclosure is schematically shown.

[0125] like Figure 5 As shown, the method of this embodiment includes operations S501 to S504.

[0126] Iteratively do the following:

[0127] In operation S501, it is determined whether the fuzzy similarity matrix satisfies an iteration stop condition;

[0128] According to an embodiment of the present disclosure, the iteration stop condition may include When the fuzzy similarity matrix meets the iteration stop condition, it means that the matrix S q It has transitivity, where S represents the fuzzy similarity matrix and q represents the number of operations on the fuzzy similarity matrix.

[0129] The operator represents the synthesis of fuzzy similarity matrices, The operation rules of the operator are as follows:

[0130] Suppose there are two data matrices R1 and R2. The composite calculation formula of R1 and R2 is shown in formula (4). The two matrices that can be composited need to satisfy the requirement that the number of columns of R1 is equal to the number of rows of R2.

[0131]

[0132] Where e represents the number of rows in the matrix R1, f represents the number of columns in the matrix R2, z represents the number of columns in the matrix R1 and the number of rows in R2, and the maximum value of z is h; the ∨ and ∧ operators are called Zadeh operators. The V operator takes the maximum value from a set of data elements, and the ∧ operator takes the minimum value from a set of data elements.

[0133] In operation S502, if the fuzzy similarity matrix does not satisfy the iteration stop condition, a square operation is performed on the fuzzy similarity matrix to generate a first fuzzy similarity matrix; and

[0134] In operation S503, the first fuzzy similarity matrix is ​​used as a fuzzy similarity matrix; and

[0135] In operation S504 , when the fuzzy similarity matrix satisfies an iteration stop condition, the fuzzy similarity matrix is ​​used as a fuzzy equivalence matrix.

[0136] According to the embodiment of the present disclosure, starting from the fuzzy similarity matrix S, the quadratic operation of the fuzzy similarity matrix S can be performed in sequence. When it first appeared When , it indicates that the matrix S q With transitivity, the matrix S q That is the fuzzy equivalent matrix t(S) of the fuzzy similarity matrix S.

[0137] Figure 6 The flowchart of clustering resources in an initial resource database based on a fuzzy equivalence matrix and generating a resource database according to an embodiment of the present disclosure is schematically shown.

[0138] like Figure 6 As shown, the method of this embodiment includes operations S601 to S602.

[0139] In operation S601 , a plurality of thresholds to be classified are obtained from a fuzzy equivalence matrix.

[0140] In operation S602, resources in the initial resource database are clustered according to each to-be-classified threshold to generate a resource database.

[0141] According to the embodiment of the present disclosure, the threshold to be classified can be first set to 1, and an element equal to 1 can be found in the fuzzy equivalence matrix. If the element t(x, y) in the fuzzy equivalence matrix is ​​1, it means that vq x and vq y If the two video resource objects belong to the same class, they can be clustered and placed in the same cluster. Otherwise, they do not belong to the same class, and the search for data objects that meet the requirements continues. Each different classification threshold corresponds to a different classification result. The value of the classification threshold is then continuously reduced, where the value of the classification threshold is selected from the fuzzy equivalence matrix. This hierarchical clustering method is a bottom-up dynamic clustering process. When the classification threshold is selected as 1, all video resources are individually classified into one class. Then, by continuously reducing the threshold, new clustering results are continuously generated. Ultimately, all elements are merged into one cluster, forming a dynamic clustering process.

[0142] Figure 7 The flowchart of determining a target cluster from multiple clusters according to an embodiment of the present disclosure is schematically shown.

[0143] like Figure 7 As shown, the method of this embodiment includes operations S701 to S703.

[0144] In operation S701 , classification description data of each cluster is obtained.

[0145] In operation S702 , statistics are calculated based on each category description data to generate multiple statistical results.

[0146] In operation S703 , the cluster corresponding to the maximum value of the statistical result is determined as the target cluster.

[0147] According to the embodiment of the present disclosure, in order to determine the optimal threshold value λ to be classified and thus obtain the best clustering result, the variance analysis method in statistics can be used to use the F statistic to obtain the optimal threshold value λ to be classified. Assume that the number of classifications of all video resources based on quality information is N, and the number of video resources in the rth classification is n. r , and the sample of the rth classification is defined as The calculation formula of F statistic is shown in formula (4):

[0148]

[0149] in It represents the average value of all video resource quality information attributes in the classification results, and is calculated as shown in formula (5), where n vq Indicates the total number of video resources in the recommended resources; It represents the average value of the video resource quality information attribute in the rth cluster. is the data center of quality information, and the calculation method is shown in formula (6).

[0150]

[0151] i represents the number of rows in the data matrix in step 1, and j represents the number of columns in the data matrix.

[0152]

[0153] c represents the number of rows in the video resource quality information matrix of the rth cluster, Represents the sum of data in each row of the quality information matrix of the video resources in the rth cluster; for arrive The Euclidean distance is calculated as shown in formula (7).

[0154]

[0155] is the sum of the quality information data of the video resources in the rth cluster To the "Quality" data center The Euclidean distance of .

[0156] The F statistic is a statistic in statistics with a degree of freedom of (N-1, n r -N) distribution function, the denominator of formula (5) represents the distance between each data object sample in the cluster, and the numerator represents the distance between each cluster after the clustering result is generated; by calculating the F statistics corresponding to different classification thresholds, the F statistics with the largest value is found. At this time, the corresponding threshold λ value is the optimal threshold that meets the conditions, and the clustering result under this threshold is the optimal clustering result.

[0157] Figure 8 The flowchart of processing statistical data matrix and generating a standardized matrix according to an embodiment of the present disclosure is schematically shown.

[0158] like Figure 8 As shown, the method of this embodiment includes operations S801 to S802.

[0159] In operation S801 , the statistical data matrix is ​​normalized to generate a first matrix.

[0160] In operation S802 , a translation range change process is performed on the first matrix to generate a normalized matrix.

[0161] The statistical data matrix is ​​processed as follows:

[0162] First, the data elements in the statistical data matrix can be transformed by translation standard deviation. The transformation formula is shown in formula (8):

[0163]

[0164] Among them, vq ij is an element in the "quality" data matrix of video resources. Its size represents the attribute value of different quality information of each video resource data. n is the number of rows in the data matrix, which represents the total number of video resources. j is the number of columns in the data matrix, which represents the dimension of quality information. represents the average value of each row element in the data matrix, as shown in formula (9), v j Represents the standard deviation of each column of data elements in the data matrix, as shown in formula (10).

[0165]

[0166]

[0167] Then, the data in the first matrix can be subjected to a translation range transformation, as shown in formula (11).

[0168]

[0169] where min 1≤i≤n {vq′ ij} represents the minimum value of each column element in the data matrix, max 1≤i≤n {vq′ ij} represents the maximum value among the elements in each column of the data matrix.

[0170] According to the embodiments of the present disclosure, the standardization of the statistical data matrix can normalize the data of different data dimensions in the statistical data matrix to [0, 1], effectively avoiding the problems caused by different numerical dimensions. Because excessively large data may lead to different orders of magnitude and have a significant impact on the final data classification results; at the same time, the standardization process can also balance the impact of video resource quality information on classification, achieving better classification results.

[0171] Figure 9 The flowchart of using the first description information and the second description information to train the resource recommendation model to be trained and generating the trained resource recommendation model according to an embodiment of the present disclosure is schematically shown.

[0172] like Figure 9 As shown, the method of this embodiment includes operations S901 to S903.

[0173] In operation S901 , the first description information and the second description information are input into a resource recommendation model to be trained, and a prediction result is output, wherein the prediction result represents a predicted click probability of a target object to be recommended for a resource corresponding to the second description information.

[0174] In operation S902 , the prediction result and label information are input into a loss function, and a loss result is output.

[0175] According to an embodiment of the present disclosure, the first description information and the second description information may be first fitted into a data matrix, and the data matrix may be used as a training sample.

[0176] According to an embodiment of the present disclosure, the resource recommendation model to be trained may be trained based on a logistic regression algorithm.

[0177] According to an embodiment of the present disclosure, the loss function can be expressed by the following formula (12).

[0178]

[0179] Among them, m represents the total number of training samples, f w represents the sigmoid function, y i Indicates label information, x i Represents training samples.

[0180] In operation S903 , the network parameters of the resource recommendation model to be trained are iteratively adjusted according to the loss result to generate a trained resource recommendation model.

[0181] According to an embodiment of the present disclosure, the network parameters may include the weight, bias, and step size of the resource recommendation model to be trained.

[0182] After obtaining the first description information, label information, and second description information, the gradient ascent linear fitting module can be first called to initialize the network parameters of the resource recommendation model to be trained. The first and second description information are then fitted into an initial parameter matrix, and the initial parameter matrix is ​​processed to generate a standardized data matrix. This data processing may include data cleaning and data standardization.

[0183] After the standardized data matrix is ​​generated, the row data in the standardized data matrix can be sequentially input into the resource recommendation model to be trained until all the row data in the standardized data matrix are traversed.

[0184] In the process of inputting the standardized data matrix into the resource recommendation model to be trained, each time a row of data is input, the training step alpha value can be dynamically calculated using formula (13).

[0185]

[0186] Among them, j and i can represent the data coordinates in the row data respectively.

[0187] After traversing all the row data in the standardized data matrix, at least one row of data can be randomly selected from the standardized data matrix, the inner product of the row data is calculated based on the updated weight matrix, and then the inner product is input into the sigmoid function to output the function value.

[0188] After the function value is calculated, the loss value can be calculated based on the function value and label information, and the network parameters of the resource recommendation model can be iteratively adjusted based on the loss value.

[0189] According to an embodiment of the present disclosure, by randomly selecting rows of data, calculating the loss value based on the row of data and label information, and adjusting the network parameters of the model based on the calculated loss value, the randomness of the data in the model training process can be increased, and the resource recommendation accuracy of the trained resource recommendation model can be improved while improving the training efficiency.

[0190] In the embodiments of the present disclosure, fuzzy clustering is used to effectively realize the clustering of resources, so that high-quality resources can be quickly recalled and the recommendation frequency of low-quality resources is reduced; in addition, the first descriptive information characterizing the user's characteristic information is introduced through the logistic regression algorithm, and the resource recommendation problem is converted into a user click-through rate estimation problem, thereby improving the effectiveness and accuracy of the recommendation results.

[0191] According to an embodiment of the present disclosure, when the resource is a video resource, the quality information includes the number of plays, likes, comments, and reposts of the video resource.

[0192] Based on the training method of the resource recommendation model, the present disclosure also provides a resource recommendation method. Figure 10 The method is described in detail.

[0193] Figure 10 The flowchart of the resource recommendation method according to an embodiment of the present disclosure is schematically shown.

[0194] like Figure 10 As shown, the method of this embodiment includes operations S1001 to S1002.

[0195] In operation S1001, description information of an object to be recommended is input into a resource recommendation model, and a prediction result array is output. The prediction result array includes a plurality of prediction probability values, each probability value representing a click probability of the object to be recommended for at least one resource. The resource recommendation model is trained using the resource recommendation model training method provided by an embodiment of the present disclosure.

[0196] In operation S1002 , resource recommendations are made to the object to be recommended based on the prediction result array.

[0197] Based on the above resource recommendation model training method, the present disclosure also provides a resource recommendation model training device. Figure 11 The device is described in detail.

[0198] Figure 11 The structural block diagram of the training device of the resource recommendation model according to the embodiment of the present disclosure is schematically shown.

[0199] like Figure 11 As shown, the resource recommendation model training device 1100 of this embodiment includes a first acquisition module 1101 , a first determination module 1102 , a second acquisition module 1103 and a training module 1104 .

[0200] The first acquisition module 1101 is used to acquire a resource database, wherein the resource database includes multiple clusters obtained by pre-clustering, and each cluster includes at least one resource. In one embodiment, the first acquisition module 1101 can be used to perform the operation S201 described above, which will not be repeated here.

[0201] The first determining module 1102 is configured to determine a target cluster from a plurality of clusters. In one embodiment, the first determining module 1102 may be configured to perform the operation S202 described above, which will not be described in detail herein.

[0202] Second acquisition module 1103 is configured to acquire first description information of a user associated with resources in the target cluster, and second description information of the resources in the target cluster, wherein the first description information represents the user's preference for the resources, and the first description information includes tag information, which represents the user's probability of selecting the resources in the target cluster. In one embodiment, second acquisition module 1103 may be configured to perform operation S203 described above, and will not be further described here.

[0203] The training module 1104 is used to train the resource recommendation model to be trained using the first description information, the tag information, and the second description information to generate a trained resource recommendation model. In one embodiment, the training module 1104 can be used to perform the operation S204 described above, which will not be repeated here.

[0204] According to an embodiment of the present disclosure, the training device 1100 for the resource recommendation model further includes a first generation module, a second generation module, a third generation module, a fourth generation module, and a fifth generation module.

[0205] A first generating module is used to generate a statistical data matrix based on the quality information of each resource in the initial resource database, wherein the quality information is obtained based on log data generated by users' operations on the resources;

[0206] The second generation module is used to process the statistical data matrix and generate a standardized matrix;

[0207] The third generating module is used to perform fuzzy similarity calculation on the data in the standardized matrix to generate a fuzzy similarity matrix;

[0208] A fourth generating module, configured to generate a fuzzy equivalence matrix based on the fuzzy similarity matrix;

[0209] The fifth generating module is used to cluster the resources in the initial resource database based on the fuzzy equivalence matrix to generate a resource database.

[0210] According to an embodiment of the present disclosure, the fifth generating module includes a first acquiring unit and a first generating unit.

[0211] A first acquiring unit is used to acquire a plurality of thresholds to be classified from a fuzzy equivalence matrix;

[0212] The first generating unit is configured to cluster the resources in the initial resource database according to each threshold to be classified, so as to generate a resource database.

[0213] According to an embodiment of the present disclosure, the fourth generating module includes a first determining unit, a first operating unit, a second determining unit, and a third determining unit.

[0214] A first determining unit, configured to determine whether the fuzzy similarity matrix satisfies an iteration stop condition;

[0215] a first operation unit, configured to perform a square operation on the fuzzy similarity matrix to generate a first fuzzy similarity matrix when the fuzzy similarity matrix does not satisfy an iteration stop condition; and

[0216] a second determining unit, configured to use the first fuzzy similarity matrix as a fuzzy similarity matrix; and

[0217] The third determining unit is configured to use the fuzzy similarity matrix as a fuzzy equivalence matrix when the fuzzy similarity matrix satisfies an iteration stopping condition.

[0218] According to an embodiment of the present disclosure, a row of data in the normalized matrix represents one resource.

[0219] According to an embodiment of the present disclosure, the third generation module includes a second operation unit and a second generation unit.

[0220] The second operation unit is used to calculate the similarity between any two rows of data in the standardized data matrix until each row of data in the standardized data matrix is ​​traversed to generate a similarity result;

[0221] The second generating unit is used to generate a fuzzy similarity matrix according to the similarity result.

[0222] According to an embodiment of the present disclosure, the second generation module includes a third generation unit and a fourth generation unit.

[0223] a third generating unit, configured to perform standardization processing on the statistical data matrix to generate a first matrix;

[0224] The fourth generating unit is used to perform translation range change processing on the first matrix to generate a standardized matrix.

[0225] According to an embodiment of the present disclosure, the first determining module 1102 includes a first acquiring unit, a fifth generating unit, and a sixth generating unit.

[0226] A first acquisition unit is used to acquire classification description data of each cluster;

[0227] a fifth generating unit, configured to calculate statistics according to each classification description data and generate multiple statistical results;

[0228] The sixth generating unit is configured to determine the cluster corresponding to the maximum value of the statistical result as the target cluster.

[0229] According to an embodiment of the present disclosure, the training module 1104 includes a first output unit, a second output unit, and a seventh generating unit.

[0230] A first output unit is configured to input the first description information and the second description information into a resource recommendation model to be trained, and output a prediction result, wherein the prediction result represents a predicted click probability of the target object to be recommended for the resource corresponding to the second description information;

[0231] The second output unit is used to input the prediction result and label information into the loss function and output the loss result;

[0232] The seventh generating unit is used to iteratively adjust the network parameters of the resource recommendation model to be trained according to the loss result to generate a trained resource recommendation model.

[0233] According to an embodiment of the present disclosure, when the resource is a video resource, the quality information includes the number of plays, likes, comments, and reposts of the video resource.

[0234] Based on the above resource recommendation method, the present disclosure also provides a resource recommendation device. Figure 12 The device is described in detail.

[0235] Figure 12 The structural block diagram of the resource recommendation device according to an embodiment of the present disclosure is schematically shown.

[0236] like Figure 12 As shown, the resource recommendation device 1200 of this embodiment includes an input module 1201 and a recommendation module 1202 .

[0237] Input module 1201 is used to input the description information of the recommended object into the resource recommendation model and output a prediction result array, where the prediction result array includes multiple predicted probability values, each probability value representing the probability of clicking the recommended object on at least one resource. The resource recommendation model is trained using the resource recommendation model training method provided in the embodiments of the present disclosure. In one embodiment, input module 1201 can be used to perform operation S1001 described above, which will not be repeated here.

[0238] The recommendation module 1202 is used to recommend resources to the recommended object based on the prediction result array. In one embodiment, the recommendation module 1202 can be used to perform the operation S1002 described above, which will not be repeated here.

[0239] According to an embodiment of the present disclosure, any multiple modules among the first acquisition module 1101, the first determination module 1102, the second acquisition module 1103, the training module 1104, the input module 1201, and the recommendation module 1202 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first acquisition module 1101, the first determination module 1102, the second acquisition module 1103, the training module 1104, the input module 1201, and the recommendation module 1202 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the first acquisition module 1101, the first determination module 1102, the second acquisition module 1103, the training module 1104, the input module 1201 and the recommendation module 1202 can be at least partially implemented as a computer program module, which can perform corresponding functions when executed.

[0240] Figure 13 The block diagram schematically shows an electronic device suitable for implementing a resource recommendation model training method and a resource recommendation method according to an embodiment of the present disclosure.

[0241] like Figure 13 As shown, the electronic device 1300 according to an embodiment of the present disclosure includes a processor 1301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1302 or a program loaded from a storage part 1308 into a random access memory (RAM) 1303. The processor 1301 may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1301 may also include an onboard memory for caching purposes. The processor 1301 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0242] Various programs and data required for the operation of the electronic device 1300 are stored in the RAM 1303. The processor 1301, the ROM 1302, and the RAM 1303 are connected to each other via a bus 1304. The processor 1301 executes the various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 1302 and / or the RAM 1303. It should be noted that the programs may also be stored in one or more memories other than the ROM 1302 and the RAM 1303. The processor 1301 may also execute the various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0243] According to an embodiment of the present disclosure, electronic device 1300 may further include an input / output (I / O) interface 1305, which is also connected to bus 1304. Electronic device 1300 may further include one or more of the following components connected to I / O interface 1305: an input portion 1306 including a keyboard, mouse, etc.; an output portion 1307 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage portion 1308 including a hard disk; and a communication portion 1309 including a network interface card such as a LAN card or modem. Communication portion 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to I / O interface 1305 as needed. Removable media 1311, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1310 as needed, so that computer programs read from the removable media can be installed into storage portion 1308 as needed.

[0244] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0245] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 1302 and / or RAM 1303 described above and / or one or more memories other than ROM 1302 and RAM 1303.

[0246] Embodiments of the present disclosure also include a computer program product, comprising a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to cause the computer system to implement the resource recommendation model training method and resource recommendation method provided in the embodiments of the present disclosure.

[0247] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the computer program is executed by the processor 1301. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0248] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal over a network medium, downloaded and installed via the communication portion 1309, and / or installed from removable media 1311. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0249] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1309 and / or installed from the removable medium 1311. When the computer program is executed by the processor 1301, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0250] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0251] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0252] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways, even if such combinations and / or couplings are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or couplings are intended to fall within the scope of this disclosure.

[0253] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for training a resource recommendation model, comprising: Acquire a resource database, wherein the resource database includes a plurality of clusters obtained through pre-clustering, and each cluster includes at least one resource; Obtaining classification description data for each of the clusters; Calculating statistics according to each of the classification description data to generate multiple statistical results; Determine the cluster corresponding to the maximum value of the statistical result as the target cluster; Obtaining first description information of a user associated with a resource in the target cluster, and second description information of the resource in the target cluster, wherein the first description information represents the user's preference for the resource, the first description information has tag information, the tag information represents the user's selection probability for the resource in the target cluster, and the second description information represents the attribute of the resource in the target cluster; and The resource recommendation model to be trained is trained using the first description information, the tag information, and the second description information to generate a trained resource recommendation model.

2. The method according to claim 1, wherein The resource database is generated by the following operations: Generating a statistical data matrix based on quality information of each resource in the initial resource database, wherein the quality information is obtained based on log data generated by users performing operations on the resources; Performing data processing on the statistical data matrix to generate a standardized matrix; Performing fuzzy similarity calculation on the data in the standardized matrix to generate a fuzzy similarity matrix; generating a fuzzy equivalence matrix based on the fuzzy similarity matrix; Based on the fuzzy equivalence matrix, the resources in the initial resource database are clustered to generate the resource database.

3. The method according to claim 2, wherein: Clustering the resources in the initial resource database based on the fuzzy equivalence matrix to generate the resource database includes: Obtaining a plurality of threshold values ​​to be classified from the fuzzy equivalence matrix; The resources in the initial resource database are clustered according to each threshold to be classified to generate the resource database.

4. The method according to claim 2, wherein: Generating a fuzzy equivalence matrix based on the fuzzy similarity matrix includes: Iteratively do the following: Determining whether the fuzzy similarity matrix satisfies an iteration stop condition; If the fuzzy similarity matrix does not satisfy the iteration stop condition, performing a square operation on the fuzzy similarity matrix to generate a first fuzzy similarity matrix; and Using the first fuzzy similarity matrix as the fuzzy similarity matrix; and When the fuzzy similarity matrix satisfies the iteration stopping condition, the fuzzy similarity matrix is ​​used as the fuzzy equivalence matrix.

5. The method according to claim 2, wherein: A row of data in the standardized matrix represents one resource; The performing fuzzy similarity calculation on the data in the standardized matrix to generate a fuzzy similarity matrix includes: Calculating the similarity between any two rows of data in the standardized matrix until every row of data in the standardized matrix is ​​traversed to generate a similarity result; The fuzzy similarity matrix is ​​generated according to the similarity result.

6. The method according to claim 2, wherein: The performing data processing on the statistical data matrix to generate a standardized matrix includes: performing standardization processing on the statistical data matrix to generate a first matrix; Perform translation range change processing on the first matrix to generate the standardized matrix.

7. The method according to claim 1, wherein the step of training a resource recommendation model to be trained using the first description information and the second description information to generate a trained resource recommendation model comprises: Inputting the first description information and the second description information into the resource recommendation model to be trained, and outputting a prediction result, wherein the prediction result represents the predicted click probability of the target object to be recommended for the resource corresponding to the second description information; Input the prediction result and the label information into a loss function, and output a loss result; Iteratively adjust the network parameters of the resource recommendation model to be trained according to the loss result to generate the trained resource recommendation model.

8. The method according to claim 2, wherein: In the case where the resource is a video resource, the quality information includes the number of plays, likes, comments, and reposts of the video resource.

9. A resource recommendation method, comprising: Inputting description information of the object to be recommended into a resource recommendation model, and outputting a prediction result array, wherein the prediction result array includes a plurality of prediction probability values, each of which represents a click probability of the object to be recommended for at least one resource, and the resource recommendation model is trained by the resource recommendation model training method according to any one of claims 1 to 8; According to the prediction result array, resource recommendations are made to the object to be recommended.

10. A training device for a resource recommendation model, comprising: A first acquisition module is configured to acquire a resource database, wherein the resource database includes a plurality of clusters obtained through pre-clustering, and each cluster includes at least one resource; A first determining module, configured to determine a target cluster from the plurality of clusters; a second acquisition module, configured to acquire first description information of a user associated with a resource in the target cluster, and second description information of the resource in the target cluster, wherein the first description information represents the user's preference for the resource, the first description information has tag information, the tag information represents the user's selection probability for the resource in the target cluster, and the second description information represents an attribute of the resource in the target cluster; and a training module, configured to train a resource recommendation model to be trained using the first description information, the label information, and the second description information, and generate a trained resource recommendation model; The first determining module includes: A first acquiring unit, configured to acquire classification description data of each cluster; a fifth generating unit, configured to calculate statistics according to each of the classification description data to generate a plurality of statistical results; The sixth generating unit is configured to determine the cluster corresponding to the maximum value of the statistical result as the target cluster.

11. A resource recommendation device, comprising: An input module, configured to input description information of an object to be recommended into a resource recommendation model and output a prediction result array, wherein the prediction result array includes a plurality of prediction probability values, each probability value representing a click probability of the object to be recommended for at least one resource, and the resource recommendation model is trained by the resource recommendation model training method according to any one of claims 1 to 8; The recommendation module is used to recommend resources to the object to be recommended based on the prediction result array.

12. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to perform the method according to any one of claims 1 to 9.

13. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to perform the method according to any one of claims 1 to 9.

14. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images