CN116128575A - Article recommendation method, apparatus, computer equipment, storage medium and program product - Google Patents

Abstract

The application relates to an item recommendation method, an item recommendation device, a computer device, a storage medium and a computer program product. The method comprises the following steps: constructing an initial undirected weight graph according to the user nodes, the object nodes corresponding to the user nodes, the relation among the user nodes and the object nodes; the user nodes comprise head nodes corresponding to head users and long-tail nodes corresponding to long-tail users; inputting the initial undirected graph into a preset graph self-encoder model, and updating the relation between long tail nodes and other nodes to generate a target undirected graph; and recommending the articles for the long-tail user according to the initial undirected unauthorized image and the target undirected unauthorized image, and generating article recommendation results. By adopting the method, the article recommendation can be more accurately performed on the long-tail user according to the constructed initial undirected graph and the target undirected graph containing more comprehensive feedback information, so that a more accurate article recommendation result is generated.

Description

Article recommendation method, apparatus, computer equipment, storage medium and program product

technical field

The present application relates to the technical field of artificial intelligence, in particular to an item recommendation method, device, computer equipment, storage medium and computer program product.

Background technique

When recommending items, various financial institutions generally need to recommend items personalizedly for different users, so as to improve the accuracy of item recommendation and user satisfaction with services.

In general, when recommending items, the targeted users may include head users and long-tail users. Among them, the top user refers to the user to be recommended with sufficient feedback information. Long-tail users refer to users to be recommended who only contain a small amount of feedback information or whose feedback information is incomplete. Therefore, for long-tail users, due to the lack of sufficient feedback information such as browsing history information and label information, when recommending items to long-tail users based on a small number of feedback information or incomplete feedback information, there is a problem of inaccurate item recommendation. .

Contents of the invention

Based on this, it is necessary to provide an item recommendation method, device, computer equipment, computer readable storage medium and computer program product capable of improving the accuracy of item recommendation in order to address the above technical problems.

In a first aspect, the present application provides an item recommendation method. The methods include:

According to the user node, the item node corresponding to the user node, the relationship between the user nodes, the relationship between the user node and the item node, construct an initial undirected and unweighted graph; the user node includes The head node corresponding to the head user and the long tail node corresponding to the long tail user;

Inputting the initial undirected and unweighted graph into a preset graph autoencoder model, updating the relationship between the long-tail node and other nodes, and generating a target undirected and unweighted graph;

According to the initial undirected and unweighted graph and the target undirected and unweighted graph, item recommendation is performed to the long-tail user, and an item recommendation result is generated.

In one of the embodiments, the construction of an initial undirected Unauthorized graphs, including:

Using the user node and the item node corresponding to the user node as nodes, using the relationship between the user nodes, the relationship between the user node and the item node as edges, and constructing an initial adjacency matrix;

Obtaining the feature vector of the user node and the feature vector of the item node, and generating a feature matrix according to the feature vector of the user node and the feature vector of the item node;

According to the initial adjacency matrix and feature matrix, the initial undirected and unweighted graph is constructed.

In one embodiment, the method also includes:

Deleting the connection edges of the head nodes in the initial undirected and unweighted graph to generate a new undirected and unweighted graph;

Inputting the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

Calculate the value of the loss function of the initial graph self-encoder model according to the new adjacency matrix, update the model parameters of the initial graph self-encoder model according to the value of the loss function, and generate a preset graph self-encoder device model.

In one of the embodiments, the initial undirected and unweighted graph is input into the preset graph autoencoder model, and the relationship between the long-tail node and other nodes is updated to generate the target undirected and unweighted graph. rights map, including:

Inputting the initial undirected and unweighted graph into the preset graph autoencoder model, predicting the connection edges to be added of the long-tail nodes, and generating a target adjacency matrix;

The target undirected and unweighted graph is generated according to the target adjacency matrix and the feature matrix.

In one embodiment, the method also includes:

Inputting the initial undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a first embedded representation matrix corresponding to the initial undirected and unweighted graph;

Inputting the target undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a second embedded representation matrix corresponding to the target undirected and unweighted graph;

Calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, and update the model parameters of the initial graph neural network model according to the value of the loss function , to generate a preset graph neural network model.

In one of the embodiments, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, item recommendation is performed to the long-tail user, and item recommendation results are generated, including:

Inputting the initial undirected and unweighted graph into a preset graph neural network model for graph convolution processing to generate a target embedding representation matrix; the preset graph neural network model is based on the initial undirected and unweighted graph and the The target undirected and unweighted graph is obtained by training;

performing item recommendation on the long-tail user according to the target embedding representation matrix, and generating the item recommendation result.

In a second aspect, the present application also provides an item recommendation device. The devices include:

An initial undirected and unweighted graph construction module, configured to construct an initial An undirected and unweighted graph; the user node includes a head node corresponding to the head user and a long tail node corresponding to the long tail user;

The target undirected and unweighted graph generation module is used to input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between the long tail node and other nodes, and generate the target undirected and unweighted graph. to the Unauthorized Graph;

The item recommendation result generating module is configured to recommend items to the long-tail users according to the initial undirected and unweighted graph and the target undirected and unweighted graph, and generate item recommendation results.

In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method in any one embodiment of the first aspect above when executing the computer program.

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer-readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method in any one embodiment of the above-mentioned first aspect are implemented.

In a fifth aspect, the present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method in any one embodiment of the first aspect above are implemented.

The above item recommendation method, device, computer equipment, storage medium, and computer program product construct an initial undirected Unweighted graph; user nodes include the head node corresponding to the head user and the long tail node corresponding to the long tail user; the initial undirected unweighted graph is input into the preset graph autoencoder model, and the long tail node and other The relationship between the nodes is updated to generate the target undirected and unweighted graph; according to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. This application can expand the imperfect feedback information of long-tail users by inputting the constructed initial undirected and unweighted graph into the preset graph autoencoder model and updating the relationship between long-tail nodes and other nodes. Thus, a target undirected and unweighted graph containing more comprehensive feedback information is generated. Afterwards, according to the constructed initial undirected and unweighted graph and the target undirected and unweighted graph containing more comprehensive feedback information, items can be more accurately recommended to long-tail users, thereby generating more accurate item recommendation results.

Description of drawings

FIG. 1 is an application environment diagram of an item recommendation method in an embodiment;

FIG. 2 is a schematic flow diagram of an item recommendation method in an embodiment;

Fig. 3 is a schematic flow chart of the initial undirected and unweighted graph construction steps in one embodiment;

FIG. 4 is a schematic structural diagram of an initial undirected and unweighted graph in an embodiment;

Fig. 5 is a schematic flow chart of the steps of generating the preset graph autoencoder model in another embodiment;

Fig. 6 is a schematic diagram of the training of the preset graph self-encoder model in one embodiment;

Fig. 7 is a schematic flow chart of the step of generating a target undirected and unweighted graph in an embodiment;

Fig. 8 is a schematic flow chart of the steps of generating a preset graph neural network model in another embodiment;

Fig. 9 is a schematic diagram of training of a preset graph neural network model in an embodiment;

FIG. 10 is a schematic flowchart of the steps of generating item recommendation results in an embodiment;

FIG. 11 is a schematic flow chart of an item recommendation method in an optional embodiment;

Fig. 12 is a schematic structural diagram of an item recommendation system in an embodiment;

Fig. 13 is a structural block diagram of an item recommendation device in an embodiment;

Figure 14 is a diagram of the internal structure of a computer device in one embodiment.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

When recommending items, various financial institutions generally need to recommend items personalizedly for different users, so as to improve the accuracy of item recommendation and user satisfaction with services.

In general, when recommending items, the targeted users may include head users and long-tail users. Among them, the top user refers to the user to be recommended with sufficient feedback information. Long-tail users refer to users to be recommended who only contain a small amount of feedback information or whose feedback information is incomplete. Therefore, for long-tail users, due to the lack of sufficient feedback information such as browsing history information and label information, when recommending items to long-tail users based on a small number of feedback information or incomplete feedback information, there is a problem of inaccurate item recommendation. .

The item recommendation method provided in the embodiment of the present application may be applied to the application environment shown in FIG. 1 . Wherein, the terminal 102 communicates with the server 104 through the network. The data storage system can store data that needs to be processed by the server 104 . The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. The server 104 constructs an initial undirected and unweighted graph according to the user nodes, item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes; the user nodes include the head corresponding to the head user node and the long-tail node corresponding to the long-tail user; the server 104 inputs the initial undirected and unweighted graph into the preset graph autoencoder model, updates the relationship between the long-tail node and other nodes, and generates the target undirected and unweighted graph. Weight graph: The server 104 recommends items to long-tail users according to the initial undirected and unweighted graph and the target undirected and unweighted graph, and generates item recommendation results. Among them, the terminal 102 can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. . Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, and the like. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

In one embodiment, as shown in FIG. 2 , a method for item recommendation is provided. The method is applied to the server 104 in FIG. 1 as an example for illustration, including the following steps:

Step 220, construct an initial undirected and unweighted graph according to user nodes, item nodes corresponding to user nodes, relationship between user nodes, and relationship between user nodes and item nodes; user nodes include head users corresponding to head users Long-tail nodes corresponding to internal nodes and long-tail users.

Wherein, a node refers to a vertex on the graph, and may include a user node and an item node, and the user node includes a head node corresponding to a head user and a long tail node corresponding to a long tail user. Top users refer to users to be recommended with sufficient feedback information. Long-tail users refer to users to be recommended who only contain a small amount of feedback information or whose feedback information is incomplete. The feedback information refers to the collected historical behavior information of the user. Exemplarily, the feedback information includes behaviors such as browsing and liking when the user browses video software, and behaviors such as clicking, saving, and adding to a shopping cart when the user browses commodity software. An undirected and unweighted graph refers to a graph in which the relationship between user nodes and the relationship between user nodes and item nodes have no direction and no weight.

Optionally, the server 104 can obtain the user's historical behavior information from the terminal 102, and determine the user, the items used by the user when performing the historical behavior, the relationship between different users, and the relationship between the user and the item according to the user's historical behavior information. , so as to determine the user node, the item node corresponding to the user node, the relationship between the user nodes, and the relationship between the user node and the item node. Afterwards, the server 104 may construct an initial undirected and unweighted graph according to user nodes, item nodes corresponding to the user nodes, relationships between user nodes, and relationships between user nodes and item nodes.

Step 240, input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between the long-tail nodes and other nodes, and generate the target undirected and unweighted graph.

Optionally, the server 104 can input the constructed initial undirected and unweighted graph into the preset graph autoencoder model for processing, for the relationship between long-tail nodes and other user nodes, or for the relationship between long-tail nodes and other user nodes The relationship between item nodes is updated, so as to expand the relationship information between long tail nodes and other nodes, and then generate the target undirected and unweighted graph. Among them, the preset graph autoencoder model (Graph Autoencoders, GAE) is a model that uses input information as an encoder to reconstruct the original graph. The target undirected and unweighted graph refers to the undirected and unweighted graph after information augmentation of the long tail nodes in the initial undirected and unweighted graph.

Step 260, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, recommend items to long-tail users, and generate item recommendation results.

Optionally, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, the server 104 may recommend items to long-tail users and generate item recommendation results. Exemplarily, the server 104 may input the initial undirected and unweighted graph and the target undirected and unweighted graph into the preset graph neural network model for processing, and generate a processing result; then perform item recommendation according to the processing result, and generate item recommendation result. The server 104 can also train the initial undirected and unweighted graph together with the target undirected and unweighted graph to obtain the preset graph neural network model, and then input the initial undirected and unweighted graph into the preset graph neural network model for processing, and generate Result; afterward, the server 104 may perform item recommendation according to the processing result, and generate an item recommendation result. Among them, the graph neural network model (graph neural network, GNN) refers to a model that uses a neural network to learn graph-structured data to meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. Optionally, in this embodiment of the application, a graph convolutional neural network model may also be used for processing, which is not limited in this application.

In the above item recommendation method, an initial undirected and unweighted graph is constructed according to user nodes, item nodes corresponding to user nodes, relationship between user nodes, and relationship between user nodes and item nodes; user nodes include The corresponding head node and the long-tail node corresponding to the long-tail user; input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between the long-tail node and other nodes, and generate the target unweighted Directed and unweighted graph: According to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. This application can expand the imperfect feedback information of long-tail users by inputting the constructed initial undirected and unweighted graph into the preset graph autoencoder model and updating the relationship between long-tail nodes and other nodes. Thus, a target undirected and unweighted graph containing more comprehensive feedback information is generated. Afterwards, according to the constructed initial undirected and unweighted graph and the target undirected and unweighted graph containing more comprehensive feedback information, it is possible to more accurately recommend personalized items to long-tail users, thereby generating more accurate item recommendation results. Ultimately, user satisfaction can be improved.

In one embodiment, as shown in Figure 3, according to the user nodes, the item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes, an initial undirected and unweighted graph is constructed, including :

In step 320, user nodes and item nodes corresponding to the user nodes are used as nodes, and relationships between user nodes and relationships between user nodes and item nodes are used as edges to construct an initial adjacency matrix.

Optionally, firstly, the server 104 may obtain the user's historical behavior information, and determine the user, the items used by the user when performing the historical behavior, the relationship between different users, and the relationship between the user and the item according to the user's historical behavior information. relation. Afterwards, the server 104 can use each user as each user node, each item used by the user when performing historical behaviors as each item node, and use the user node and the item node corresponding to the user node as nodes, and the Relationships, relationships between user nodes and item nodes are used as edges, and an initial adjacency matrix is constructed based on each node and each edge. Among them, the initial adjacency matrix can be expressed as A, and A∈{0,1} ^n×n , n represents the total number of nodes. Each element A _ij in the initial adjacency matrix A indicates whether there is a connected edge between node i and node j. If there is a connecting edge between node i and node j, the element value of A _ij is 1; if there is no connecting edge between node i and node j, then the element value of A _ij is 0.

Step 340, acquire the feature vector of the user node and the feature vector of the item node, and generate a feature matrix according to the feature vector of the user node and the feature vector of the item node.

n表示节点总数，f表示特征向量的维数，矩阵的每一行x_i表示节点i的f维特征向量。独热编码可以将离散的特征信息数值化，以便于服务器运算。例如，独热编码可以采用两位二进制数表示性别特征，若第一位表示“是否为女性”，且第二位表示“是否为男性”，则女性可以独热编码为“10”，而男性独热编码为“01”。Optionally, the server 104 may obtain the user's historical behavior information, and obtain the feature information of each user node and each item node according to the user's historical behavior information and the determined user nodes and item nodes. Afterwards, the server 104 can one-hot encode the feature information of each user node and the feature information of each item node into a feature vector of each user node and a feature vector of each item node, and based on the feature vector of each user node and the feature vector of each item node The eigenvectors generate the eigenmatrix X. where the feature matrix can be denoted as X, and

n represents the total number of nodes, f represents the dimension of the feature vector, and each row x _i of the matrix represents the f-dimensional feature vector of node i. One-hot encoding can digitize discrete feature information to facilitate server calculations. For example, one-hot encoding can use two binary numbers to represent gender characteristics. If the first digit indicates "whether it is female" and the second digit indicates "whether it is male", then female can be one-hot encoded as "10", while male One-hot encoding is "01".

Step 360, construct an initial undirected and unweighted graph according to the initial adjacency matrix and feature matrix.

Optionally, the server 104 may construct an initial undirected and unweighted graph according to the initial adjacency matrix A and the feature matrix X. Among them, an undirected and unweighted graph refers to a graph in which the relationship between user nodes and the relationship between user nodes and item nodes have no direction and no weight value, that is, each edge in the graph has no direction and no A graph of weights. The initial undirected and unweighted graph can be represented as G, and G=(A,X), A represents the initial adjacency matrix, and X represents the feature matrix. Exemplarily, as shown in FIG. 4 , FIG. 4 is a schematic structural diagram of an initial undirected and unweighted graph in an embodiment. The nodes include the head node v ₀ , the long tail node v ₁ , the item node v ₂ , the item node v ₃ , the item node v ₄ and the item node v ₅ , each node includes the feature vector corresponding to each node, and the edge includes the head node The following relationship between v ₀ and long tail node v ₁ , the sharing relationship between head node v ₀ and item node v ₂ , the like relationship between head node v ₀ and item node v ₃ , the head node Like relationship between v ₀ and item node v ₄ and like relationship between head node v ₀ and item node v ₅ .

In this embodiment, the user node and the item node corresponding to the user node are used as nodes, and the relationship between the user nodes and the relationship between the user node and the item node are used as edges to construct an initial adjacency matrix; obtain the feature vector of the user node , The eigenvector of the item node, according to the eigenvector of the user node and the eigenvector of the item node, generate a feature matrix; according to the initial adjacency matrix and feature matrix, construct an initial undirected and unweighted graph. Therefore, the initial undirected and unweighted graph is constructed through the initial adjacency matrix and feature matrix, and the relationship between users and the relationship between users and items can be obtained more conveniently and accurately.

In one embodiment, as shown in FIG. 5 , an item recommendation method is provided, further comprising:

Step 520, delete the connecting edges of the head nodes in the initial undirected and unweighted graph, and generate a new undirected and unweighted graph.

Exemplarily, as shown in FIG. 6 , FIG. 6 is a schematic diagram of training a preset graph autoencoder model in an embodiment. Optionally, the server 104 may delete the connection edges of the head nodes in the initial undirected and unweighted graph, and generate a new adjacency matrix A ₁ after edge deletion, thereby generating a new undirected and weightless graph G ₁ after edge deletion . Wherein, G ₁ =(A ₁ ,X). When deleting the connection edges of the head node in the initial undirected and unweighted graph, any one or more connection edges connected to the head node can be randomly deleted. Of course, the present application does not limit this.

Step 540: Input the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix.

Optionally, as shown in FIG. 6, the server 104 can input the new undirected and unweighted graph _G1 after edge deletion into the initial graph self-encoder model for processing, that is, the server 104 can input the new undirected and unweighted graph G ₁ is input into the initial graph autoencoder model, and predicts the connection edge to be added to the head node whose connection edge has been deleted, and generates a new adjacency matrix A ₂ after the predicted edge is added.

Step 560, calculate the value of the loss function of the initial GAS model according to the new adjacency matrix, update the model parameters of the initial GAS model according to the value of the loss function, and generate a preset GAS model.

Optionally, as shown in FIG. 6 , the server 104 can calculate the value of the loss function Loss _p of the initial graph autoencoder model according to the new adjacency matrix _A2 after prediction and edge addition, and according to the loss function of the initial graph autoencoder model The value of Loss _p updates the model parameter Θ of the initial graph autoencoder model until the value of the loss function Loss _p of the initial graph autoencoder model is minimum, that is, until the new adjacency matrix A ₂ is adjacent to the initial adjacency When the gap of matrix A is the smallest, use the model parameters corresponding to the loss function Loss _p of the initial graph autoencoder model at this time as the target model parameters of the initial graph autoencoder model, thereby generating a trained preset graph autoencoder model . Wherein, Θ=(L ₁ , L ₂ ) is the model parameter of the initial graph autoencoder model. The calculation formula of the loss function Loss _p of the initial image self-encoder model is shown in formula (1), formula (2), and formula (3).

H＝A ₁ relu(A ₁ XL ₁ )L ₂ (1)

A ₂ =sigmoid(HH ^T ) (2)

Among them, H represents the intermediate model parameters of the initial graph self-encoder model; A ₁ represents the new adjacency matrix after edge deletion, X represents the feature matrix; relu() represents the first nonlinear activation function, defined as h(x)=max (0,x); L ₁ represents the first model parameter of the initial graph autoencoder model; L ₂ represents the second model parameter of the initial graph autoencoder model; A ₂ represents the new adjacency matrix after predicting and adding edges; Sigmoid( ) represents the second nonlinear activation function, defined as h(x)=1/(1+e ^-x ); ^HT represents the transposition of the intermediate model parameters of the initial graph autoencoder model; Loss _p represents the initial graph autoencoder The loss function of the device model; A represents the initial adjacency matrix; n represents the total number of nodes, and i and j represent different nodes. In the embodiment of the present application, the relu activation function may also be replaced by an ELU activation function or a Leaky Relu activation function.

In this embodiment, the connection edges of the head nodes in the initial undirected and unweighted graph are deleted to generate a new undirected and unweighted graph; the new undirected and unweighted graph is input into the initial graph autoencoder model to perform Processing to generate a new adjacency matrix; calculate the value of the loss function of the initial graph self-encoder model according to the new adjacency matrix, update the model parameters of the initial graph self-encoder model according to the value of the loss function, and generate a preset graph self-encoder device model. In this embodiment, the preset graph autoencoder model is jointly trained by the initial adjacency matrix, the new adjacency matrix after edge deletion, and the new adjacency matrix after prediction and edge addition, so as to generate a more accurate preset graph autoencoder model.

In one embodiment, as shown in Figure 7, the initial undirected and unweighted graph is input into the preset graph autoencoder model, and the relationship between long-tail nodes and other nodes is updated to generate the target undirected and unweighted Figures, including:

Step 720: Input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix.

Optionally, after the training of the preset graph self-encoder model is completed, the server 104 can input the initial undirected and unweighted graph G into the preset graph self-encoder model to predict the connection edges to be added to the long-tail nodes, so as to Update the relationship between the long-tail node and other user nodes, or the relationship between the long-tail node and other item nodes, that is, expand the relationship information between the long-tail node and other nodes to generate the target adjacency matrix . Among them, the preset graph autoencoder model (Graph Autoencoders, GAE) is a model that uses input information as an encoder to reconstruct the original graph. The target adjacency matrix can be expressed as A'. The calculation formula of the target adjacency matrix A' is shown in formula (4), formula (5), formula (6) and formula (7).

H＝Arelu(AXL ₁ )L ₂ (4)

AP＝softmax(HH ^T ) (5)

AM＝Bernoulli(AP) (6)

A ^′ ＝Clamp(AM+A) (7)

Among them, AP represents the first intermediate variable of the target adjacency matrix; softmax() represents the normalized exponential function; AM represents the second intermediate variable of the target adjacency matrix; Bernoulli() represents Bernoulli sampling; Clamp() represents the matrix element Values are truncated to [0,1]; A' represents the target adjacency matrix.

Step 740: Generate a target undirected and unweighted graph according to the target adjacency matrix and feature matrix.

Optionally, the server 104 can generate the target undirected and unweighted graph G' according to the target adjacency matrix A' and the feature matrix X. Among them, G'=(A',X). Among them, the target adjacency matrix refers to the adjacency matrix after expanding the relationship information between the long tail node and other nodes. The target undirected and unweighted graph refers to the undirected and unweighted graph after the information expansion of the relationship between the long-tail nodes and other nodes in the initial undirected and unweighted graph.

In this embodiment, the initial undirected and unweighted graph is input into the preset graph autoencoder model, and the connection edges to be added to the long-tail nodes are predicted to generate the target adjacency matrix; according to the target adjacency matrix and feature matrix, the target An undirected and unweighted graph can expand the imperfect feedback information of long-tail users, thereby generating a target undirected and unweighted graph that contains more comprehensive feedback information.

In one embodiment, as shown in FIG. 8 , an item recommendation method is provided, which further includes:

Step 820: Input the initial undirected and unweighted graph into the initial graph neural network model to perform graph convolution processing to generate a first embedded representation matrix corresponding to the initial undirected and unweighted graph.

且n表示节点总数，c表示嵌入表征维数，矩阵的每一行z_i表示节点i的c维嵌入表征，且c远小于f。Φ＝(W₁,W₂)表示初始图神经网络模型的模型参数。第一嵌入表征矩阵Z₁的计算公式如公式(8)所示。Exemplarily, as shown in FIG. 9 , FIG. 9 is a schematic diagram of training a preset graph neural network model in an embodiment. Optionally, the server 104 may input the initial undirected and unweighted graph Z into the initial graph neural network model to perform graph convolution processing to generate the first embedded representation matrix corresponding to the initial undirected and unweighted graph G. Among them, the embedded representation matrix can be expressed as

And n represents the total number of nodes, c represents the embedded representation dimension, each row z _i of the matrix represents the c-dimensional embedded representation of node i, and c is much smaller than f. Φ=(W ₁ , W ₂ ) represents the model parameters of the initial graph neural network model. The calculation formula of the first embedded representation matrix Z ₁ is shown in formula (8).

Z ₁ ＝Arelu(AXW ₁ )W ₂ (8)

Among them, Z ₁ represents the first embedded representation matrix; A represents the initial adjacency matrix; relu() represents the first nonlinear activation function, defined as h(x)=max(0,x); W ₁ represents the initial graph neural network model The first model parameter of W ; W ₂ represents the second model parameter of the initial graph neural network model.

Step 840: Input the target undirected and unweighted graph into the initial graph neural network model to perform graph convolution processing to generate a second embedded representation matrix corresponding to the target undirected and unweighted graph.

Optionally, as shown in FIG. 9 , the server 104 may input the target undirected and unweighted graph G' into the initial graph neural network model for graph convolution processing, and generate the second embedding corresponding to the target undirected and unweighted graph G' representation matrix. Wherein, the calculation formula of the second embedded representation matrix Z ₂ is shown in formula (9).

Z ₂ ＝A ^′ relu(A ^′ XW ₁ )W ₂ (9)

Among them, Z ₂ represents the second embedded representation matrix; A' represents the target adjacency matrix; relu() represents the first nonlinear activation function, defined as h(x)=max(0,x); W ₁ represents the initial graph neural network The first model parameter of the model; W ₂ represents the second model parameter of the initial graph neural network model.

Step 860, calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, update the model parameters of the initial graph neural network model according to the value of the loss function, and generate a preset graph neural network Model.

Optionally, as shown in FIG. 9, the server 104 can calculate the value of the loss function Loss _q of the initial graph neural network model according to the first embedded representation matrix _Z1 and the second embedded representation matrix _Z2 , and according to the initial graph neural network model The value of the loss function Loss _q of the initial graph neural network model is updated to the model parameter Φ until the loss function Loss _{q of the initial graph neural network model is the smallest, and the model corresponding to the loss function Loss q of} the initial graph neural network model at this time is used The parameters are used as the target model parameters of the initial graph neural network model to generate a trained preset graph neural network model. Among them, the calculation formula of the loss function Loss _q of the initial graph neural network model is shown in formula (10).

Loss _q = CrossEntropy(Z ₁ ,Y)+CrossEntropy(Z ₂ ,Y) (10)

Among them, Loss _q represents the loss function of the initial graph neural network model; CrossEntropy() represents the cross-entropy loss function, which is a commonly used deep learning model loss function and is used to measure the similarity between two probability distributions ; Z ₁ represents the first embedded representation matrix; Y is the known and invariable node label; Z ₂ represents the second embedded representation matrix.

In this embodiment, the initial undirected and unweighted graph is input into the initial graph neural network model for graph convolution processing, and the first embedded representation matrix corresponding to the initial undirected and unweighted graph is generated; the target undirected and unweighted graph is input into Perform graph convolution processing in the initial graph neural network model to generate the second embedded representation matrix corresponding to the target undirected and unweighted graph; calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix , according to the value of the loss function, the model parameters of the initial graph neural network model are updated to generate a preset graph neural network model. This embodiment uses the initial undirected and unweighted graph and the target undirected and unweighted graph G' after the expansion of long-tail node information to jointly train the preset graph neural network model. Compared with the traditional method, only the initial undirected and unweighted graph G is used. The training method of the preset graph neural network model, obviously, the preset graph neural network model training method in the embodiment of the present application is more accurate, therefore, the embodiment of the present application can train and generate a more accurate preset graph neural network model.

In one embodiment, as shown in FIG. 10 , according to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated, including:

Step 1020, input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedding representation matrix; the preset graph neural network model is based on the initial undirected and unweighted graph and the target undirected and unweighted graph Figure obtained from training.

Optionally, after the training of the preset graph neural network model is completed, the server 104 may input the initial undirected and unweighted graph G into the preset graph neural network model for graph convolution processing, thereby generating the target embedding representation matrix Z. Wherein, the preset graph neural network model is a graph neural network model obtained by training based on an initial undirected and unweighted graph and a target undirected and unweighted graph. A graph neural network model (graph neural network, GNN) refers to a model that uses a neural network to learn graph-structured data to meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. Optionally, in this embodiment of the application, a graph convolutional neural network model may also be used for processing, which is not limited in this application. Each row _zi of the target embedding representation matrix Z represents the embedding representation vector of node i.

Step 1040, recommend items to long-tail users according to the target embedding representation matrix, and generate item recommendation results.

Optionally, first, the server 104 may obtain the long-tail user number i to be recommended, that is, determine the long-tail user i to be recommended. Secondly, according to the target embedding representation matrix Z _, the server 104 can calculate the _cosine similarity degree, and generate a cosine similarity list (s1, s2, s3,...,sm). Among them, m represents the length of the preset item list to be recommended, and m represents the length of the cosine similarity list. sj represents the cosine similarity between the embedded representation of long-tail user i and the embedded representation of item j; cosine similarity is also called cosine similarity, and cosine similarity is to evaluate two vectors by calculating the cosine value of the angle between them of similarity. In the embodiment of the present application, the cosine similarity may also be replaced by Pearson correlation coefficient or Jaccard similarity coefficient (Jaccard similarity coefficient). Afterwards, the cosine similarities in the cosine similarity list (s1, s2, s3,...,sm) are sorted to generate a cosine similarity sorted result. Then, select the items with the highest preset number (k) of cosine similarity corresponding to the long-tail user i from the cosine similarity sorting results, and the preset number (k) of the highest cosine similarity corresponding to the long-tail user i ( k) items are determined as target items. Then, output the target item corresponding to the long-tail user i to the long-tail user i, that is, output the item recommendation result of the long-tail user i. Wherein, the embodiment of the present application does not limit the preset number (k).

In this embodiment, since the embodiment of the application has trained a more accurate preset graph neural network model based on the initial undirected and unweighted graph and the target undirected and unweighted graph, the initial undirected and unweighted graph is input into the more accurate By performing graph convolution processing on the preset graph neural network model, a more accurate target embedding representation matrix can be generated. Afterwards, according to the more accurate target embedding representation matrix to recommend items to long-tail users, more accurate item recommendation results can be generated.

In an optional embodiment, as shown in FIG. 11 , a method for item recommendation is provided. The method is applied to the server 104 in FIG. 1 as an example for illustration, including the following steps:

Step 1102, using user nodes and item nodes corresponding to the user nodes as nodes, and using the relationship between user nodes and the relationship between user nodes and item nodes as edges to construct an initial adjacency matrix;

Step 1104, obtaining the feature vector of the user node and the feature vector of the item node, and generating a feature matrix according to the feature vector of the user node and the feature vector of the item node;

Step 1106, construct an initial undirected and unweighted graph according to the initial adjacency matrix and feature matrix; user nodes include head nodes corresponding to head users and long-tail nodes corresponding to long-tail users;

Step 1108, delete the connection edge of the head node in the initial undirected and unweighted graph, and generate a new undirected and unweighted graph;

Step 1110, inputting the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

Step 1112, calculate the value of the loss function of the initial graph autoencoder model according to the new adjacency matrix, update the model parameters of the initial graph autoencoder model according to the value of the loss function, and generate a preset graph autoencoder model;

Step 1114, input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix;

Step 1116, generate a target undirected and unweighted graph according to the target adjacency matrix and feature matrix;

Step 1118, input the initial undirected and unweighted graph into the initial graph neural network model for graph convolution processing, and generate the first embedded representation matrix corresponding to the initial undirected and unweighted graph;

Step 1120, input the target undirected and unweighted graph into the initial graph neural network model to perform graph convolution processing, and generate a second embedded representation matrix corresponding to the target undirected and unweighted graph;

Step 1122, calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, update the model parameters of the initial graph neural network model according to the value of the loss function, and generate a preset graph neural network Model;

Step 1124, input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedding representation matrix; the preset graph neural network model is based on the initial undirected and unweighted graph and the target undirected and unweighted graph The graph is obtained through training;

Step 1126, recommend items to long-tail users according to the target embedding representation matrix, and generate item recommendation results.

The above item recommendation method constructs an initial undirected and unweighted graph according to the user nodes, the item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes; the user nodes include The head node and the long-tail node corresponding to the long-tail user; input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between the long-tail node and other nodes, and generate the target undirected Unweighted graph: According to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. This application can expand the imperfect feedback information of long-tail users by inputting the constructed initial undirected and unweighted graph into the preset graph autoencoder model and updating the relationship between long-tail nodes and other nodes. Thus, a target undirected and unweighted graph containing more comprehensive feedback information is generated. Afterwards, according to the constructed initial undirected and unweighted graph and the target undirected and unweighted graph containing more comprehensive feedback information, items can be more accurately recommended to long-tail users, thereby generating more accurate item recommendation results.

It should be understood that although the steps in the flow charts involved in the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least some of the steps in the flow charts involved in the above embodiments may include multiple steps or stages, and these steps or stages are not necessarily executed at the same time, but may be executed at different times, The execution order of these steps or stages is not necessarily performed sequentially, but may be executed in turn or alternately with other steps or at least a part of steps or stages in other steps.

In an optional embodiment, as shown in FIG. 12 , an item recommendation system 1200 is provided. The system runs on the server 104 in FIG. 1 as an example for illustration. The item recommendation system 1200 includes a data processing module 1220 , a long-tail node enhancement module 1240, a graph neural network module 1260, and a recommendation module 1280.

Wherein, the data processing module 1220 is used to generate an initial undirected and unweighted graph according to the information of each user and each item. The data processing module 1220 includes: modeling users and items as nodes, modeling user feedback behaviors on items and interaction behaviors between users as edges to obtain an initial adjacency matrix A; one-hot encoding of feature information of users and items as The f-dimensional feature vector obtains the feature matrix X; according to the adjacency matrix A and the feature matrix X, an initial undirected and unweighted graph G is constructed.

Wherein, the long-tail node enhancement module 1240 is configured to perform long-tail node enhancement based on the initial undirected and unweighted graph, and generate an enhanced target undirected and unweighted graph. The long-tail node enhancement module 1240 includes: performing random edge deletion processing on the head nodes in the initial undirected and unweighted graph G; updating the preset graph autoencoder model according to the undirected and unweighted graph training after random edge deletion; The graph self-encoder model is set to predict the connection edges added to the long-tail nodes, and the target undirected and unweighted graph G' is output.

Wherein, the graph neural network module 1260 is used to predict the embedding representation matrix according to the initial undirected and unweighted graph and the target undirected and unweighted graph, and generate the target embedding representation matrix. The graph neural network module 1260 includes: training and updating the preset graph neural network model according to the initial undirected and unweighted graph G and the target undirected and unweighted graph G'; inputting the initial undirected and unweighted graph G into the preset graph neural network model After processing, the target embedding representation matrix Z is output.

Among them, the recommendation module 1280 is used to recommend personalized items to long-tail users according to the user number i corresponding to the long-tail users and the target embedding representation matrix. The recommendation module 1280 includes: calculating the cosine similarity between the specified long-tail user i and the embedded representations of all items; sorting the items according to the cosine similarity, and selecting the top k items as personalized recommendation results for output.

Based on the same inventive concept, an embodiment of the present application further provides an item recommendation device for implementing the above-mentioned item recommendation method. The solution to the problem provided by the device is similar to the implementation described in the above method, so the specific limitations in one or more embodiments of the item recommendation device provided below can refer to the above-mentioned limitations on the item recommendation method, I won't repeat them here.

In one embodiment, as shown in FIG. 13 , an item recommendation device 1300 is provided, including: an initial undirected and unweighted graph construction module 1320, a target undirected and unweighted graph generation module 1340, and an item recommendation result generation module 1360, in:

The initial undirected and unweighted graph construction module 1320 is used to construct an initial undirected and unweighted graph according to the user nodes, item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes; The nodes include head nodes corresponding to head users and long tail nodes corresponding to long tail users.

The target undirected and unweighted graph generation module 1340 is used to input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between long tail nodes and other nodes, and generate the target undirected and unweighted graph picture.

The item recommendation result generating module 1360 is configured to recommend items to long-tail users according to the initial undirected and unweighted graph and the target undirected and unweighted graph, and generate item recommendation results.

In one embodiment, the initial undirected and unweighted graph construction module 1320 includes:

The initial adjacency matrix construction unit is used to use user nodes and item nodes corresponding to the user nodes as nodes, and use the relationship between user nodes and the relationship between user nodes and item nodes as edges to construct an initial adjacency matrix;

A feature matrix generation unit is used to obtain the feature vector of the user node and the feature vector of the item node, and generate a feature matrix according to the feature vector of the user node and the feature vector of the item node;

The initial undirected and unweighted graph construction unit is used to construct the initial undirected and unweighted graph according to the initial adjacency matrix and the feature matrix.

In one embodiment, the item recommendation device 1300 further includes:

The new undirected and unweighted graph generation module is used to delete the connection edges of the head nodes in the initial undirected and unweighted graph to generate a new undirected and unweighted graph;

The new adjacency matrix generation module is used to input the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

The preset graph autoencoder model generation module is used to calculate the value of the loss function of the initial graph autoencoder model according to the new adjacency matrix, update the model parameters of the initial graph autoencoder model according to the value of the loss function, and generate the predicted Build a graph autoencoder model.

In one embodiment, the target undirected and unweighted graph generation module 1340 includes:

The target adjacency matrix generation unit is used to input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix;

The target undirected and unweighted graph generation unit is used to generate the target undirected and unweighted graph according to the target adjacency matrix and the feature matrix.

In one embodiment, the item recommendation device 1300 further includes:

The first embedding representation matrix generation module is used to input the initial undirected and unweighted graph into the initial graph neural network model to perform graph convolution processing, and generate the first embedding representation matrix corresponding to the initial undirected and unweighted graph;

The second embedding representation matrix generation module is used to input the target undirected and unweighted graph into the initial graph neural network model to perform graph convolution processing, and generate the second embedding representation matrix corresponding to the target undirected and unweighted graph;

The preset graph neural network model generation module is used to calculate the value of the loss function of the initial graph neural network model according to the first embedding representation matrix and the second embedding representation matrix, and calculate the model parameters of the initial graph neural network model according to the value of the loss function Update, generate preset graph neural network model.

In one embodiment, the item recommendation result generation module 1360 includes:

The target embedding representation matrix generation unit is used to input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedded representation matrix; the preset graph neural network model is based on the initial undirected and unweighted graph And the target undirected and unweighted graph is obtained from training;

The item recommendation result generating unit is configured to recommend items to long-tail users according to the target embedding representation matrix, and generate item recommendation results.

Each module in the above item recommending device can be fully or partially realized by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, and can also be stored in the memory of the computer device in the form of software, so that the processor can invoke and execute the corresponding operations of the above-mentioned modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure may be as shown in FIG. 14 . The computer device includes a processor, a memory, an input/output interface (Input/Output, I/O for short), and a communication interface. Wherein, the processor, the memory and the input/output interface are connected through the system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs and databases. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store item recommendation data. The input/output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an item recommendation method is realized.

Those skilled in the art can understand that the structure shown in Figure 14 is only a block diagram of a partial structure related to the solution of this application, and does not constitute a limitation on the computer equipment on which the solution of this application is applied. The specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, a computer program is stored in the memory, and the processor implements the following steps when executing the computer program:

According to the user node, the item node corresponding to the user node, the relationship between the user nodes, and the relationship between the user node and the item node, the initial undirected and unweighted graph is constructed; the user node includes the head node corresponding to the head user and The long-tail node corresponding to the long-tail user;

Input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between long-tail nodes and other nodes, and generate the target undirected and unweighted graph;

According to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated.

In one embodiment, an initial undirected and unweighted graph is constructed according to the user nodes, the item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes. When the processor executes the computer program, it also Implement the following steps:

Use user nodes and item nodes corresponding to user nodes as nodes, and use the relationship between user nodes and the relationship between user nodes and item nodes as edges to construct an initial adjacency matrix;

Obtain the feature vector of the user node and the feature vector of the item node, and generate a feature matrix according to the feature vector of the user node and the feature vector of the item node;

According to the initial adjacency matrix and feature matrix, the initial undirected and unweighted graph is constructed.

In one embodiment, the following steps are also implemented when the processor executes the computer program:

Delete the connection edges of the head nodes in the initial undirected and unweighted graph to generate a new undirected and unweighted graph;

Input the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

Calculate the value of the loss function of the initial graph autoencoder model according to the new adjacency matrix, update the model parameters of the initial graph autoencoder model according to the value of the loss function, and generate a preset graph autoencoder model.

In one embodiment, the initial undirected and unweighted graph is input into the preset graph autoencoder model, and the relationship between the long tail node and other nodes is updated to generate the target undirected and unweighted graph, and the processor executes the computer The procedure also implements the following steps:

Input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix;

According to the target adjacency matrix and feature matrix, the target undirected and unweighted graph is generated.

In one embodiment, the following steps are also implemented when the processor executes the computer program:

Inputting the initial undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a first embedded representation matrix corresponding to the initial undirected and unweighted graph;

Inputting the target undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a second embedded representation matrix corresponding to the target undirected and unweighted graph;

Calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, update the model parameters of the initial graph neural network model according to the value of the loss function, and generate a preset graph neural network model.

In one embodiment, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. When the processor executes the computer program, the following steps are also implemented:

Input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedding representation matrix; the preset graph neural network model is trained based on the initial undirected and unweighted graph and the target undirected and unweighted graph income;

According to the target embedding representation matrix, items are recommended for long-tail users, and item recommendation results are generated.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

According to the user node, the item node corresponding to the user node, the relationship between the user nodes, and the relationship between the user node and the item node, the initial undirected and unweighted graph is constructed; the user node includes the head node corresponding to the head user and The long-tail node corresponding to the long-tail user;

Input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between long-tail nodes and other nodes, and generate the target undirected and unweighted graph;

According to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated.

In one embodiment, an initial undirected and unweighted graph is constructed according to the user nodes, the item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes. When the computer program is executed by the processor Also implement the following steps:

Use user nodes and item nodes corresponding to user nodes as nodes, and use the relationship between user nodes and the relationship between user nodes and item nodes as edges to construct an initial adjacency matrix;

Obtain the feature vector of the user node and the feature vector of the item node, and generate a feature matrix according to the feature vector of the user node and the feature vector of the item node;

According to the initial adjacency matrix and feature matrix, the initial undirected and unweighted graph is constructed.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented:

Delete the connection edges of the head nodes in the initial undirected and unweighted graph to generate a new undirected and unweighted graph;

Input the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

Calculate the value of the loss function of the initial graph autoencoder model according to the new adjacency matrix, update the model parameters of the initial graph autoencoder model according to the value of the loss function, and generate a preset graph autoencoder model.

In one embodiment, the initial undirected and unweighted graph is input into the preset graph autoencoder model, the relationship between the long-tail nodes and other nodes is updated, and the target undirected and unweighted graph is generated, and the computer program is processed The following steps are also implemented when the controller executes:

Input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix;

According to the target adjacency matrix and feature matrix, the target undirected and unweighted graph is generated.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented:

Inputting the initial undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a first embedded representation matrix corresponding to the initial undirected and unweighted graph;

Inputting the target undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a second embedded representation matrix corresponding to the target undirected and unweighted graph;

Calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, update the model parameters of the initial graph neural network model according to the value of the loss function, and generate a preset graph neural network model.

In one embodiment, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. When the computer program is executed by the processor, the following steps are also implemented:

Input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedding representation matrix; the preset graph neural network model is trained based on the initial undirected and unweighted graph and the target undirected and unweighted graph income;

According to the target embedding representation matrix, items are recommended for long-tail users, and item recommendation results are generated.

In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

According to the user node, the item node corresponding to the user node, the relationship between the user nodes, and the relationship between the user node and the item node, the initial undirected and unweighted graph is constructed; the user node includes the head node corresponding to the head user and The long-tail node corresponding to the long-tail user;

Input the initial undirected and unweighted graph into the preset graph autoencoder model, update the relationship between long-tail nodes and other nodes, and generate the target undirected and unweighted graph;

According to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated.

In one embodiment, an initial undirected and unweighted graph is constructed according to the user nodes, the item nodes corresponding to the user nodes, the relationship between user nodes, and the relationship between user nodes and item nodes. When the computer program is executed by the processor Also implement the following steps:

Use user nodes and item nodes corresponding to user nodes as nodes, and use the relationship between user nodes and the relationship between user nodes and item nodes as edges to construct an initial adjacency matrix;

Obtain the feature vector of the user node and the feature vector of the item node, and generate a feature matrix according to the feature vector of the user node and the feature vector of the item node;

According to the initial adjacency matrix and feature matrix, the initial undirected and unweighted graph is constructed.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented:

Delete the connection edges of the head nodes in the initial undirected and unweighted graph to generate a new undirected and unweighted graph;

Input the new undirected and unweighted graph into the initial graph autoencoder model for processing to generate a new adjacency matrix;

Calculate the value of the loss function of the initial graph autoencoder model according to the new adjacency matrix, update the model parameters of the initial graph autoencoder model according to the value of the loss function, and generate a preset graph autoencoder model.

In one embodiment, the initial undirected and unweighted graph is input into the preset graph autoencoder model, the relationship between the long-tail nodes and other nodes is updated, and the target undirected and unweighted graph is generated, and the computer program is processed The following steps are also implemented when the controller executes:

Input the initial undirected and unweighted graph into the preset graph autoencoder model, predict the connection edges to be added to the long-tail nodes, and generate the target adjacency matrix;

According to the target adjacency matrix and feature matrix, the target undirected and unweighted graph is generated.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented:

Inputting the initial undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a first embedded representation matrix corresponding to the initial undirected and unweighted graph;

Inputting the target undirected and unweighted graph into the initial graph neural network model for graph convolution processing to generate a second embedded representation matrix corresponding to the target undirected and unweighted graph;

Calculate the value of the loss function of the initial graph neural network model according to the first embedded representation matrix and the second embedded representation matrix, update the model parameters of the initial graph neural network model according to the value of the loss function, and generate a preset graph neural network model.

In one embodiment, according to the initial undirected and unweighted graph and the target undirected and unweighted graph, items are recommended for long-tail users, and item recommendation results are generated. When the computer program is executed by the processor, the following steps are also implemented:

Input the initial undirected and unweighted graph into the preset graph neural network model for graph convolution processing to generate the target embedding representation matrix; the preset graph neural network model is trained based on the initial undirected and unweighted graph and the target undirected and unweighted graph income;

According to the target embedding representation matrix, items are recommended for long-tail users, and item recommendation results are generated.

It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all It is information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

Those of ordinary skill in the art can understand that realizing all or part of the processes in the methods of the above embodiments can be completed by instructing related hardware through computer programs, and the computer programs can be stored in a non-volatile computer-readable storage medium , when the computer program is executed, it may include the procedures of the embodiments of the above-mentioned methods. Wherein, any reference to storage, database or other media used in the various embodiments provided in the present application may include at least one of non-volatile and volatile storage. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive variable memory (ReRAM), magnetic variable memory (Magnetoresistive Random Access Memory, MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (Phase Change Memory, PCM), graphene memory, etc. The volatile memory may include random access memory (Random Access Memory, RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the various embodiments provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a blockchain-based distributed database, etc., but is not limited thereto. The processors involved in the various embodiments provided by this application can be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, data processing logic devices based on quantum computing, etc., and are not limited to this.

The technical features of the above embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be It is considered to be within the range described in this specification.

The above examples only express several implementation modes of the present application, and the description thereof is relatively specific and detailed, but should not be construed as limiting the patent scope of the present application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be determined by the appended claims.

Claims (10)

1. A method of recommending items, the method comprising:

constructing an initial undirected unowned graph according to user nodes, object nodes corresponding to the user nodes, the relation among the user nodes and the object nodes; the user nodes comprise head nodes corresponding to head users and long-tail nodes corresponding to long-tail users;

Inputting the initial undirected graph into a preset graph self-encoder model, and updating the relation between the long tail node and other nodes to generate a target undirected graph;

and recommending the long-tail user according to the initial undirected graph and the target undirected graph to generate an article recommendation result.

2. The method of claim 1, wherein constructing an initial undirected graph based on user nodes, item nodes corresponding to the user nodes, relationships between the user nodes, and relationships between the user nodes and the item nodes comprises:

taking user nodes and object nodes corresponding to the user nodes as nodes, and taking the relation between the user nodes and the object nodes as edges to construct an initial adjacency matrix;

acquiring the feature vector of the user node and the feature vector of the object node, and generating a feature matrix according to the feature vector of the user node and the feature vector of the object node;

and constructing an initial undirected weight graph according to the initial adjacency matrix and the feature matrix.

3. The method according to claim 2, wherein the method further comprises:

deleting the connecting edges of the head nodes in the initial undirected graph to generate a new undirected graph;

inputting the new undirected weight map into an initial map self-encoder model for processing to generate a new adjacency matrix;

and calculating the value of a loss function of the initial graph self-encoder model according to the new adjacency matrix, and updating the model parameters of the initial graph self-encoder model according to the value of the loss function to generate a preset graph self-encoder model.

4. A method according to claim 2 or 3, wherein the inputting the initial undirected graph into a preset graph self-encoder model updates the relationship between the long tail node and other nodes to generate a target undirected graph, and comprises:

inputting the initial undirected weight-free graph into a preset graph self-encoder model, and predicting the connecting edges to be added of the long tail nodes to generate a target adjacent matrix;

and generating the target undirected unowned graph according to the target adjacency matrix and the feature matrix.

5. A method according to any one of claims 1-3, characterized in that the method further comprises:

inputting the initial undirected graph into an initial graph neural network model to perform graph convolution processing, and generating a first embedded characterization matrix corresponding to the initial undirected graph;

inputting the target undirected graph into an initial graph neural network model to perform graph convolution processing, and generating a second embedded characterization matrix corresponding to the target undirected graph;

calculating the value of a loss function of the initial graph neural network model according to the first embedded characterization matrix and the second embedded characterization matrix, and updating model parameters of the initial graph neural network model according to the value of the loss function to generate a preset graph neural network model.

6. The method of claim 5, wherein the recommending the item to the long-tailed user based on the initial undirected weight graph and the target undirected weight graph, generating an item recommendation result, comprises:

inputting the initial undirected unauthorized graph into a preset graph neural network model to carry out graph convolution processing to generate a target embedded characterization matrix; the preset graph neural network model is trained based on the initial undirected graph and the target undirected graph;

And recommending the long-tail user with the article according to the target embedded characterization matrix, and generating the article recommendation result.

7. An item recommendation device, the device comprising:

the initial undirected weight map construction module is used for constructing an initial undirected weight map according to user nodes, object nodes corresponding to the user nodes, the relation among the user nodes and the object nodes; the user nodes comprise head nodes corresponding to head users and long-tail nodes corresponding to long-tail users;

the target undirected graph generating module is used for inputting the initial undirected graph into a preset graph self-encoder model, updating the relation between the long tail node and other nodes, and generating a target undirected graph;

and the article recommendation result generation module is used for recommending the article to the long-tail user according to the initial undirected graph and the target undirected graph to generate an article recommendation result.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.

9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

Images