Graph integrated neural network-based citation network article classification and recommendation method

Abstract

The invention discloses a quotation network article classification and recommendation method based on a graph integration neural network, and the method comprises the following steps: carrying out the structure enhancement of an original quotation network, carrying out the feature enhancement of the initial features of the original quotation network, and obtaining a sub-quotation network and new features; training base classifiers based on the sub-quotation network and the new features and executing deep integration, executing integration when the data stream passes through each layer of base classifier, taking an integration result as input of a next layer, and adjusting the base classifiers in training based on neighbor information to obtain a trained graph integration neural network model; inputting to-be-classified citation network articles into the trained graph integrated neural network model to obtain the probability that the articles are predicted as corresponding categories, and classifying and sorting the articles based on the obtained probability. According to the method, the base classifiers are trained based on the sub quotation networks and the new features, and deep integration is executed, so that adjacent information among the base classifiers is mutually influenced, and the classification performance of articles and the effectiveness of recommendation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of citation network article classification and recommendation, and in particular to a citation network article classification and recommendation method based on a graph integrated neural network. Background Art

[0002] Graph neural networks are currently the most effective method for processing graph data related tasks. Graph downstream tasks include node classification, link prediction and graph classification. The node classification task aims to predict and classify the entities (nodes) that make up the graph, such as predicting which field a node (article) in the citation network belongs to. This can save a lot of manpower and reduce the human classification errors caused by a lot of work. According to the number of nodes available for training, node classification can be divided into unsupervised, semi-supervised and fully supervised node classification. Learning, predicting and classifying entities (nodes) in citation networks (graphs) is a basic and important part of graph learning, which can effectively improve the analysis and processing efficiency of citation networks in real life, such as article classification and recommendation.

[0003] For citation networks, ensemble methods can improve the classification performance of graph neural networks (GNNs) for node (article) classification tasks. Existing ensemble methods in citation networks form a strong classifier by combining a set of trained base classifiers, that is, the final output of the combined base classifiers is used for prediction. However, such ensemble methods cannot effectively promote many popular GNNs, such as graph convolutional networks, graph attention networks, graph isomorphism networks, etc., to perform well under heterogeneous conditions, which limits their applicability. In addition, such methods only combine the final outputs of the base classifiers, ignoring the hierarchical nature of GNNs that aggregate information layer by layer, which makes it impossible for the base classifiers to interact during neighbor aggregation during training, that is, during training, the articles of one base classifier cannot interact with the articles of other base classifiers. Therefore, before the final output is formed, the important immediate neighbor information cannot have a positive impact on other base classifiers. These two problems are called: low applicability and shallow ensemble. Effectively solving these two problems can improve the classification performance of GNNs in citation networks and achieve better article classification and article recommendation. Summary of the invention

[0004] In order to overcome the defects and shortcomings of the prior art, the present invention provides a citation network article classification and recommendation method based on graph ensemble neural network. The present invention designs a structure enhancement method to generate a set of graphs for training different base classifiers, and designs feature enhancement to weaken the errors caused by initial features. The present invention introduces the integration operation into the neighborhood aggregation period of the training process, so that the important adjacent information between the base classifiers can influence each other, thereby improving the article classification performance, and then improving the effectiveness of article recommendation.

[0005] The second object of the present invention is to provide a citation network article classification and recommendation system based on graph integrated neural network;

[0006] A third object of the present invention is to provide a computer readable storage medium;

[0007] A fourth object of the present invention is to provide a computer device.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] The present invention provides a method for classifying and recommending citation network articles based on a graph integrated neural network, comprising the following steps:

[0010] Perform structural enhancement on the original citation network and feature enhancement on its initial features to obtain sub-citation networks and new features respectively;

[0011] Based on the sub-citation network and the new features, the base classifiers are trained and deep integration is performed. When the data stream passes through each layer of the base classifier, the integration result is used as the input of the next layer. The base classifiers are adjusted during training based on the neighbor information to obtain the trained graph integration neural network model.

[0012] The citation network articles to be classified are input into the trained graph ensemble neural network model to obtain the probability that the articles are predicted to be of the corresponding category, and the articles are classified and sorted based on the obtained probability.

[0013] As a preferred technical solution, the structural enhancement of the original citation network specifically includes:

[0014] On the basis of retaining all articles in the original citation network, Bernoulli sampling is performed on the citation relationships between articles to obtain a sub-citation network of the original citation network.

[0015] As a preferred technical solution, the feature enhancement of the initial features specifically includes:

[0016] The feature mean of each type of labeled nodes is used as the category vector, and the category vector is added to the articles of each category according to the set probability.

[0017] As a preferred technical solution, a single-layer graph convolutional network is used as the base classifier of the graph ensemble neural network, which is expressed as:

[0018]

[0019] Among them, σ is the activation function, W l is the learnable parameter of layer l, is the formalized adjacency matrix, And D ii =∑u A iu , A is the adjacency matrix of the citation network, I is the identity matrix, D is the degree matrix, and D ii Represents the value of the i-th row and i-th column of matrix D.

[0020] As a preferred technical solution, the residual connection form of the graph convolutional network is as follows:

[0021]

[0022] Among them, H l-1 is the feature matrix of layer l-1.

[0023] As a preferred technical solution, a base classifier is trained based on the sub-citation network and new features and deep integration is performed, specifically including:

[0024] The number of generated sub-citation networks is equal to the number of base classifiers in each layer of the graph ensemble neural network;

[0025] Set the integration mode to integrate the C base classifiers of each layer, expressed as:

[0026]

[0027] Where σ is the activation function, is the output of the cth base classifier in the lth layer, α is the residual coefficient, W l,c is the cth learnable parameter of the lth layer, is the formalized adjacency matrix of the c-th sub-citation network, H l-1 is the feature matrix of the l-1 layer.

[0028] As a preferred technical solution, the base classifier is trained based on the sub-citation network and the new features and deep integration is performed. The loss function is expressed as:

[0029] L GEN =L cro +γL con

[0030]

[0031] Among them, L GEN represents the loss function of GEN, L cro represents the supervised cross entropy loss, L von represents the unsupervised consistency regularization loss, and γ represents the balance L cro and L con The coefficient of Y L represents the set of articles in the labeled training set, represents the transpose of the label vector of node r, n is the number of articles in the citation network, Is to change Z through sharpening techniques i Got it.

[0032] As a preferred technical solution, the citation network articles to be classified are input into the trained graph ensemble neural network model to obtain the probability that the articles are predicted to be of the corresponding category, which is specifically expressed as:

[0033]

[0034] Where T is the temperature, a parameter used to control the degree of category distribution, Q is the number of categories of articles in the citation network, and Z ij It is the value of the i-th row and j-th column of the model output, that is, the probability that article i is predicted to be of category j.

[0035] The present invention also provides a citation network article classification and recommendation system based on graph integrated neural network, which is used to implement the above-mentioned citation network article classification and recommendation method based on graph integrated neural network, and the system includes: a structure enhancement module, a feature enhancement module, a training module and a prediction module;

[0036] The structure enhancement module is used to perform structure enhancement on the original citation network;

[0037] The feature enhancement module is used to enhance the initial features of the original citation network to obtain a sub-citation network and new features respectively;

[0038] The training module is used to train the base classifier based on the sub-citation network and the new features and perform deep integration. When the data stream passes through each layer of the base classifier, the integration result is used as the input of the next layer. The base classifier is adjusted during training based on the neighbor information to obtain a trained graph integration neural network model;

[0039] The prediction module is used to input the citation network articles to be classified into the trained graph ensemble neural network model to obtain the probability that the articles are predicted to be of the corresponding category, and classify and sort the articles based on the obtained probabilities.

[0040] The present invention also provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the above-mentioned citation network article classification and recommendation method based on graph integrated neural network.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] (1) Based on the new structural enhancement technology, the present invention performs multiple Bernoulli sampling on the edges (citation relationships between articles) in the citation network according to a given probability. Since the original citation network contains few noise citations (edges), the sampled sub-citation network contains even fewer noise edges. The set of citation networks obtained by structural enhancement is reasonable and different, and is used to train a set of base classifiers. Instead of using the same citation network to train different base classifiers, the effectiveness of the base classifier is ensured while improving the diversity of the base classifier, which is more in line with the idea of ​​ensemble learning.

[0043] (2) The present invention is based on a new feature enhancement technology to reduce the error caused by the initial features of the citation network. The initial features are usually the representation of the keywords of the article in the word bag. That is, in the real-world citation network, there are some articles of the same type whose initial features are very different or completely different, and the initial features of articles of different types are very similar or completely the same. This is inconsistent with the cognition of the classification problem and will cause certain interference to the model training. The present invention calculates the category features of each type of article, and concatenates the category features of each article after the initial features according to the set probability to obtain new features, and then constitutes the new initial features of the citation network. Feature enhancement focuses on the problem that the initial features may be inconsistent with the cognition of the classification problem, and then alleviates this problem through category features and category matrices.

[0044] (3) The present invention extends the previous graph integration that only focuses on the output of the base classifier to each layer in the training. Based on deep integration, it fully utilizes important neighbor information and then adjusts the base classifier during training to obtain a better base classifier that is more suitable for integration. Ultimately, it improves the article classification performance and the effectiveness of article recommendation within an acceptable training time. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the method for classifying and recommending citation network articles based on graph integrated neural network of the present invention;

[0046] Figure 2 It is a schematic diagram of the architecture of the graph integrated neural network model of the present invention;

[0047] Figure 3 It is a flowchart of the feature enhancement algorithm AugmFeat of Algorithm 1 of the present invention;

[0048] Figure 4 It is a flow chart of the structure enhancement algorithm AugmStru of Algorithm 2 of the present invention;

[0049] Figure 5 It is a flowchart of the integrated neural network training process of the algorithm 3 of the present invention; Figure 6 Schematic diagram of the comparison of article classification accuracy in three public citation networks. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment provides a citation network article classification and recommendation method based on a graph integrated neural network, comprising the following steps:

[0053] S1: Perform structural enhancement on the original citation network and feature enhancement on its initial features to obtain a set of sub-citation networks and new features respectively;

[0054] like Figure 3 As shown in Algorithm 1, feature enhancement is as follows: first, the category vector is calculated, that is, the feature mean of each category of labeled nodes is taken as the category vector, and then the category vector is added to the articles of each category according to the set probability, so that the features between the same category are more similar, that is, their keywords are more similar.

[0055] like Figure 4 As shown in Figure 2, in Algorithm 2, structural enhancement is to perform Bernoulli sampling on the citation relationships between articles on the basis of retaining all articles in the original citation network, that is, retaining some original citation relationships while deleting some citation relationships. Then the new network obtained must be a subnetwork of the original citation network.

[0056] S2: Train a set of base classifiers using a set of sub-citation networks and new features, and perform deep ensembles so that the base classifiers can interact during training;

[0057] The previous shallow integration that only focused on the output of the base classifier has been expanded to a deep integration that is integrated during training, and the errors that may be caused by the initial features are alleviated through feature enhancement, and the graphs used to train different base classifiers are generated through structural enhancement. The entire model for article classification and recommendation consists of a certain number of layers, each of which contains a certain number of base classifiers, such as Figure 2 As shown, when the data stream passes through each layer of base classifiers, integration is performed, and the integration result is used as the input of the next layer. During training, the final loss is obtained based on a series of integration results. Therefore, when backpropagation adjusts parameters, the base classifiers can influence each other. In addition to the cross entropy loss function, the present invention adds a consistency regularization loss to the graph ensemble neural network for article classification and recommendation to assist in model training. After the model training is completed, the network to be classified and recommended and its initial features are input into the model to obtain a feature matrix ( Figure 2According to the matrix, articles in the network can be classified and recommended.

[0058] like Figure 2 As shown in Figure 1, the graph ensemble neural network used for article classification and recommendation in citation networks is divided into two main stages, namely the graph data enhancement stage and the graph ensemble training stage. The former includes graph feature enhancement and graph structure enhancement, and the latter extends the shallow integration that only focuses on the output of the base classifier to the deep integration in training.

[0059] Specifically, the graph integrated neural network proposed in this embodiment is abbreviated as GEN, the articles in the citation network are called nodes, the representation of the keywords of the articles in the word bag is called a feature vector, and the matrix composed of the feature vectors of all articles is called a feature matrix. The input of GEN is the original citation network and its initial feature matrix, and the output is a matrix that can be used to predict node classification. Its graph feature enhancement makes the features of nodes of the same type more similar through category feature vectors and category feature matrices. The graph structure enhancement of GEN generates a set of different sub-citation networks from the input citation network for training base classifiers to ensure that the base classifiers are good but different. GEN extends shallow integration to deep integration in training. The deep integration is spread over every message passing layer in the graph neural network, so that the base classifiers can influence and correct each other through important neighbor information during the training process to produce better models;

[0060] In this embodiment, the edge matching rate h is the ratio of the edges in the class to the total edges in the graph, which is often used to determine whether the graph is matched and to help determine the applicable scope of certain models. In the present invention, it is used to calculate the proportion of citation relationships in the citation network, thereby helping to determine whether the network is matched. The calculation method is as follows (Formula 1):

[0061]

[0062] Among them, u and v represent articles, Y u is the label vector of article u, and E is the edge set of the graph, that is, the set of citation relationships. The present invention uses a single-layer graph convolutional network as the base classifier of GEN, which is in the following form (

[0063] Formula 2):

[0064]

[0065] Among them, σ is the activation function, W l is the learnable parameter of layer l, H l-1 is the feature matrix of the l-1 layer, is a formalized adjacency matrix, which is calculated as follows (Formula 3):

[0066]

[0067] in, And D ii =∑ u A iu , A is the adjacency matrix of the citation network, I is the identity matrix, D is the degree matrix, and D ii Represents the value of the i-th row and i-th column of matrix D. DropEdge is a simple variant of graph convolutional network, which slows down the convergence speed by discarding some edges, thereby alleviating the over-smoothing problem. Its form is as follows (Formula 4):

[0068]

[0069] in, It is a formalized adjacency matrix after discarding some edges. The structural enhancement of GEN is similar to it. This embodiment also imitates the residual connection in the graph convolutional network and designs a similar residual connection for GEN to improve its applicability and depth and increase the impact of initial features and feature enhancement. The residual connection in the graph convolutional network is as follows (Formula 5):

[0070]

[0071] In this embodiment, the graph data enhancement stage of GEN consists of graph feature enhancement and graph structure enhancement, such as Figure 3 and Figure 4 Algorithms 1AugmFeat and 2AugmStru in the algorithm. Algorithm 1 consists of two large loops with a small loop embedded in it. The first large loop is used to calculate the category feature vector, that is, lines 3 to 11 of the algorithm, and the second large loop is used to enhance the features by probability, that is, lines 13 to 22 of the algorithm. Algorithm 2 consists of two large loops. The first large loop embeds a small loop to generate a set of graphs, that is, lines 1 to 8 of the algorithm. The second large loop adds all known intra-class edges to the generated graph, that is, lines 10 to 17 of the algorithm. Algorithm 3GEN is the end-to-end data processing, model training and output process of GEN. The training process in lines 9-16 mentions 3 formulas, which are important parts of the form and training of GEN;

[0072] When the original citation network and its initial features are output to GEN, the structural enhancement and feature enhancement of the graph data enhancement stage of GEN will generate a set of sub-citation networks for training base classifiers and a new feature matrix respectively. This stage is the data processing process of the graph integration training stage. The number of sub-citation networks generated should be equal to the number of base classifiers in each layer of GNE, and the number of integrations performed is equal to the depth (number of layers) of GEN. Therefore, for a GEN with L layers, there are L×C base classifiers, that is, there are C base classifiers in each layer, so the graph data enhancement stage also needs to generate C additional different citation networks. The present invention selects the average value similar to the voting method as the integration method to integrate the C base classifiers of each layer, in the following form (Formula 6):

[0073]

[0074] Where σ is the activation function, is the output of the cth base classifier in the lth layer. This embodiment designs a residual connection to obtain information from the previous layer to maintain the effect of feature enhancement, then The calculation method of is as follows (Formula 7):

[0075]

[0076] Where α is the residual coefficient, W l,c is the cth learnable parameter of the lth layer, is the formalized adjacency matrix of the c-th sub-citation network. In addition to Formula 6 and Formula 7, Figure 5 As shown, Formula 8 is also mentioned in Algorithm 3, that is, the loss calculation of GEN, which consists of two parts, namely the supervised cross entropy loss and the unsupervised consistency regularization loss, which is calculated as follows (Formula 8):

[0077] L GEN =L cro +γL con

[0078] Where L GEN represents the loss function of GEN, L cto represents the supervised cross entropy loss, L con represents the unsupervised consistency regularization loss, and γ is used to balance L cro and L von The coefficient of L cro The calculation of is based on labeled nodes, so it is called supervised, and it is calculated as follows (Formula 9):

[0079]

[0080] where Y Lrepresents the set of articles in the labeled training set, Represents the transpose of the label vector of node γ. The cross entropy loss function is often used as the loss function for classification problems. For graph neural network models that perform data augmentation to generate multiple graphs or features, the consistency regularization loss function L con It is an effective means to balance multiple graphs or features. con The present invention applies GEN to the classification and recommendation problems on the citation network. The coefficient γ is added to the cross entropy loss function to obtain the loss function of GEN, L con The calculation method of is as follows (Formula 10):

[0081]

[0082] where n is the number of articles in the citation network, Is to change Z through sharpening techniques i Got it, is the probability that node i is predicted as category j. It can be found that L con The calculation involves all articles, including unlabeled articles. The calculation method of is as follows (Formula 11):

[0083]

[0084] where T is the temperature, a parameter used to control the degree of category distribution, Q is the number of categories of articles in the citation network, and Z ij is the value of the i-th row and j-th column of the model output, that is, the probability that article i is predicted to be of category j. is the processed Z ij .

[0085] This example compares the GEN proposed in this invention with 15 publicly published benchmark models on three public citation networks to verify the advancedness of the GEN. This example uses the classification accuracy of the model on the articles in the citation network as the comparison standard. The comparison results are as follows: Figure 6 shown. Figure 6 It shows that GEN has achieved the best article classification results in three public citation networks, surpassing the fifteen benchmark models used for comparison in this invention, including early space-based and spectrum-based graph neural network models, recent space-based and polynomial-based spectrum graph neural network models, etc. The comparative test results verify the advanced nature of GEN in article classification. This embodiment can make a simple transformation of GEN to make it suitable for article recommendation, that is, articles of the same category should be recommended to readers.

[0086] In summary, the present invention extends the shallow integration that only focuses on the output of the base classifier to the deep integration in training, so that the base classifiers can interact during training, and important neighbor information can affect other classifiers. In order to train good and different base classifiers, the present invention uses a single-layer graph convolutional network as the base classifier, and designs a structural enhancement technology to generate a set of different graphs from the source graph, which are used to train the base classifier. In order to weaken the errors that may be caused by the initial features (the initial features of the same type of nodes are not similar or even completely different), the present invention designs feature enhancement, which improves the similarity of the features of the same type of nodes through category vectors and matrices, weakens the errors caused by the initial features, and designs residual connections to maintain this effect in the deep layer of the model.

[0087] Example 2

[0088] This embodiment provides a citation network article classification and recommendation system based on graph integrated neural network, which is used in the citation network article classification and recommendation method based on graph integrated neural network in the above embodiment 1. The system includes: a structure enhancement module, a feature enhancement module, a training module and a prediction module;

[0089] In this embodiment, the structure enhancement module is used to perform structure enhancement on the original citation network;

[0090] In this embodiment, the feature enhancement module is used to enhance the initial features of the original citation network to obtain a sub-citation network and new features respectively;

[0091] In this embodiment, the training module is used to train the base classifier based on the sub-citation network and the new features and perform deep integration. When the data stream passes through each layer of the base classifier, the integration result is used as the input of the next layer. The base classifier is adjusted during training based on the neighbor information to obtain a trained graph integration neural network model.

[0092] In this embodiment, the prediction module is used to input the citation network articles to be classified into the trained graph integration neural network model to obtain the probability that the articles are predicted to be of the corresponding category, and classify and sort the articles based on the obtained probability.

[0093] Example 3

[0094] This embodiment provides a storage medium, which may be a storage medium such as ROM, RAM, disk, or CD. The storage medium stores one or more programs. When the program is executed by a processor, the citation network article classification and recommendation method based on a graph integrated neural network of embodiment 1 is implemented.

[0095] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A citation network article classification and recommendation method based on graph ensemble neural network, characterized in that: The steps include: Perform structural enhancement on the original citation network and feature enhancement on its initial features to obtain sub-citation networks and new features respectively; Based on the sub-citation network and the new features, the base classifier is trained and deep integration is performed. When the data stream passes through each layer of the base classifier, the integration result is used as the input of the next layer. The base classifier is adjusted during training based on the neighbor information to obtain the trained graph integration neural network model. The citation network articles to be classified are input into the trained graph ensemble neural network model to obtain the probability that the articles are predicted to be of the corresponding category, and the articles are classified and sorted based on the obtained probability.

2. The method for classifying and recommending citation network articles based on graph integrated neural network according to claim 1, characterized in that: The structural enhancement of the original citation network specifically includes: On the basis of retaining all articles in the original citation network, Bernoulli sampling is performed on the citation relationships between articles to obtain a sub-citation network of the original citation network.

3. The method for citing network article classification and recommendation based on graph integrated neural network according to claim 1, characterized in that: The feature enhancement of the initial features specifically includes: The feature mean of each type of labeled nodes is used as the category vector, and the category vector is added to the articles of each category according to the set probability.

4. The method for classifying and recommending citation network articles based on graph integrated neural network according to claim 1, characterized in that: A single-layer graph convolutional network is used as the base classifier of the graph ensemble neural network, which is expressed as: Among them, σ is the activation function, W l is the learnable parameter of layer l, is the formalized adjacency matrix, And D ii =∑ u A iu , A is the adjacency matrix of the citation network, I is the identity matrix, D is the degree matrix, and D ii Represents the value of the i-th row and i-th column of matrix D.

5. The method for classifying and recommending citation network articles based on graph integrated neural network according to claim 4, characterized in that: The residual connection form of the graph convolutional network is as follows: Among them, H l-1 is the feature matrix of the l-1 layer.

6. The method for citing network article classification and recommendation based on graph integrated neural network according to claim 1, characterized in that: Based on the sub-citation network and the new features, the base classifiers are trained and deep ensembles are performed, including: The number of generated sub-citation networks is equal to the number of base classifiers in each layer of the graph ensemble neural network; Set the integration mode to integrate the C base classifiers of each layer, expressed as: Where σ is the activation function, is the output of the cth base classifier in the lth layer, α is the residual coefficient, W l,c is the cth learnable parameter of the lth layer, is the formalized adjacency matrix of the c-th sub-citation network, H l-1 is the feature matrix of the l-1 layer.

7. The method for citing network article classification and recommendation based on graph integrated neural network according to claim 1, characterized in that: The base classifier is trained based on the sub-citation network and the new features and deep integration is performed. The loss function is expressed as: THE GEN =L cro +γL con Among them, L GEN represents the loss function of GEN, L cro represents the supervised cross entropy loss, L con represents the unsupervised consistency regularization loss, and γ represents the balance L cro and L con The coefficient of Y L represents the set of articles in the labeled training set, represents the transpose of the label vector of node r, n is the number of articles in the citation network, Is to change Z through sharpening techniques i Got it.

8. The method for citing network article classification and recommendation based on graph integrated neural network according to claim 7, characterized in that: The citation network articles to be classified are input into the trained graph ensemble neural network model to obtain the probability that the article is predicted to be of the corresponding category, which is specifically expressed as: Where T is the temperature, a parameter used to control the degree of category distribution, Q is the number of categories of articles in the citation network, and Z ij It is the value of the i-th row and j-th column of the model output, that is, the probability that article i is predicted to be of category j.

9. A citation network article classification and recommendation system based on graph integrated neural network, characterized in that: Used to implement the citation network article classification and recommendation method based on graph integrated neural network as described in any one of claims 1 to 8, the system includes: a structure enhancement module, a feature enhancement module, a training module and a prediction module; The structure enhancement module is used to perform structure enhancement on the original citation network; The feature enhancement module is used to enhance the initial features of the original citation network to obtain a sub-citation network and new features respectively; The training module is used to train the base classifier based on the sub-citation network and the new features and perform deep integration. When the data stream passes through each layer of the base classifier, the integration result is used as the input of the next layer. The base classifier is adjusted during training based on the neighbor information to obtain a trained graph integration neural network model; The prediction module is used to input the citation network articles to be classified into the trained graph integration neural network model, obtain the probability that the articles are predicted to be of the corresponding category, and classify and sort the articles based on the obtained probabilities.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the method for classifying and recommending citation network articles based on a graph integrated neural network as described in any one of claims 1 to 7 is implemented.

Images