Citation network node classification method and classification system fusing GCN and multi-classifier
By integrating GCN and multi-classifiers, and utilizing edge weights and feature matrix updates, feature filtering and amplification, and fuzzy integral fusion, the problem of global information capture and aggregation in graph-structured data by deep learning models is solved, thereby improving the accuracy and stability of node classification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENYANG NORMAL UNIV
- Filing Date
- 2024-08-19
- Publication Date
- 2026-05-29
AI Technical Summary
Existing deep learning models have limitations in capturing and aggregating global information in graph-structured data, making it difficult to fully utilize the edge weights and attribute information between nodes, resulting in insufficient accuracy and stability in node classification.
The method of integrating GCN and multi-classifiers improves the efficiency and accuracy of feature information utilization by calculating edge weights, updating the feature matrix, filtering and amplifying features, and fusing the results of multiple classifiers through fuzzy integrals.
It significantly improves the accuracy and stability of node classification, enhances the precision and recall of classification results, and provides an innovative and efficient solution.
Smart Images
Figure BDA0004999833300000031 
Figure BDA0004999833300000032 
Figure BDA0004999833300000042
Abstract
Description
Technical Field
[0001] This invention relates to the fields of graph networks and deep learning technology, and mainly to a citation network node classification method and system that integrates GCN and multi-classifiers. The aim is to improve the classification accuracy and stability of citation network nodes by integrating feature information from different levels and dimensions. Background Technology
[0002] With the significant progress in graph data research, improving the accuracy of node classification has become a hot topic in academia and industry. Researchers have explored the characteristics and structure of graph data in depth, proposing a variety of innovative models and algorithms, which have driven the development of practical problem-solving and artificial intelligence technologies. Traditional machine learning methods, such as support vector machines, decision trees, and random forests, typically rely on numerical feature representations and corresponding labels from existing literature for training. These methods are insufficient when dealing with complex graph-structured data.
[0003] In recent years, deep learning-based node classification methods have become more prevalent due to their superior performance in handling graph-structured data. Graph Convolutional Neural Networks (GCNs) and their variants, such as Graph Attention Networks (GATs) and GraphSAGE, significantly improve the performance of classification tasks by learning feature representations of nodes through deep network structures. These methods utilize convolution operations on the graph Laplacian matrix and the aggregation of neighbor node features to update node representations, enhancing the model's ability to capture both local and global information.
[0004] While existing deep learning models have made progress in node classification, they still have limitations in capturing global information and aggregating information. Therefore, proposing a more efficient aggregation strategy and graph convolution operation to fully utilize the edge weights and attribute information between nodes, thereby improving classification accuracy and the model's generalization ability, has become an urgent problem to be solved. Summary of the Invention
[0005] In view of this, the present invention provides a citation network node classification method and system that integrates GCN and multi-classifiers to solve the problems existing in the prior art.
[0006] This invention proposes a citation network node classification method that integrates GCN and multi-classifier, comprising:
[0007] S1: The knowledge citation structure based on academic fields regards the entire citation network as an undirected graph G that includes node attributes and topology. Each academic paper is regarded as a node in the network, and the citation relationship between documents is regarded as the edge between nodes.
[0008] S2: Calculate the adjacency matrix and characteristic matrix X of the undirected graph G, where, N is the number of nodes, C is the number of features for each node, C features represent whether C words in the feature word dictionary appear in the citation, and the i-th row of the feature matrix X represents the feature vector of the i-th node.
[0009] S3: Initialize the edge weight matrix of the undirected graph G, and calculate the edge weights of the undirected graph G based on the feature matrix;
[0010] S4: Aggregate the edge weights calculated in S3 to the nodes at both ends of the edge, and update the feature matrix X;
[0011] S5: Filter and amplify the features in the updated feature matrix obtained in S4;
[0012] S6: Input the adjacency matrix obtained in S2 and the feature matrix obtained in S5 into n classifiers respectively to obtain n first classification result matrices M corresponding one-to-one with the n classifiers, where n is greater than 1. The value in the i-th row and j-th column of the first classification result matrix M corresponding to each classifier represents the predicted value of the i-th node belonging to the j-th category.
[0013] Among them, the n classifiers all use the same GCN model structure, and each GCN model is trained independently to obtain n trained GCN models;
[0014] S7: The n first classification result matrices M obtained in S6 are fused using fuzzy integral fusion to obtain the final classification result matrix Out, where, N is the number of nodes, Y is the number of categories, and the value in the i-th row and j-th column of the final classification result matrix Out represents the predicted value of the i-th node belonging to the j-th category;
[0015] S8: Select the category corresponding to the largest value in each row from the final classification result matrix obtained from S7 as the classification result of the corresponding node, thus realizing the classification of the paper.
[0016] Preferably, in S3, the formula for calculating the edge weight w(d1,d2) between any two nodes d1 and d2 in the undirected graph G is:
[0017]
[0018] In the formula: F1 and F2 are the feature word sets of nodes d1 and d2, respectively; |F1∩F2| represents the size of the intersection of the feature word sets of nodes d1 and d2; and |F1∪F2| represents the size of the union of the feature word sets of nodes d1 and d2.
[0019] Further optimization, in S4, the feature matrix X is updated using the following formula:
[0020]
[0021] In the formula, Let X represent the i-th row of the updated feature matrix X, N(i) represent the set of neighboring nodes of the i-th node, and w(i,j) represent the edge weight between the i-th node and the j-th node. j This represents the j-th row of the feature matrix X before the update.
[0022] Further optimization, in S5, the filtering and amplification steps are as follows:
[0023] First, calculate the variance of each feature;
[0024] Then, define a threshold θ var Features with variance greater than the threshold are retained.
[0025] Finally, it will be greater than the preset threshold θ f The eigenvalues are increased by a corresponding multiple based on the importance index to amplify the more prominent features.
[0026] Further optimization, in step S5, the amplified features are also standardized to have a distribution with a mean of 0 and a standard deviation of 1.
[0027] Further optimization involves obtaining n trained GCN models in S6 through different random initializations and randomization of the input data order.
[0028] Further optimization involves using Sugeno fuzzy integrals for fusion in S7.
[0029] Further optimization, in S7, fusing the n first classification result matrices M through fuzzy integrals includes the following steps:
[0030] S71: For each category, extract the corresponding columns from the n first classification result matrices M corresponding to the n classifiers to form Y second classification result matrices Z that correspond one-to-one with each category, where... N is the number of nodes, and n represents the number of classifiers. The method to obtain the second classification result matrix Z(k) corresponding to the kth category is as follows: extract the kth column from the first classification result matrix M corresponding to each classifier to form the second classification result matrix Z(k) corresponding to the kth category.
[0031] S72: Taking the second classification result matrix Z corresponding to each category as input, the Sugeno fuzzy integral function is used to calculate the fusion result for each category. The formula for calculating the fusion result of the kth category is as follows:
[0032]
[0033] Where Fused(k) represents the fusion result of the k-th category, μ(i) is the fuzzy measure, representing the weight or confidence corresponding to the i-th classifier, and Z(k) i This represents the i-th column in the second classification result matrix Z(k) corresponding to the k-th category;
[0034] S73: Combine the fusion results of each category to obtain the final classification result matrix.
[0035] This invention also proposes a citation network node classification system that integrates GCN and multi-classifiers, for performing the aforementioned citation network node classification method that integrates GCN and multi-classifiers.
[0036] This invention provides a citation network node classification method and system that integrates GCN and multi-classifiers. It re-integrates feature information from different levels and dimensions, fusing, filtering, and selecting low-dimensional edge weights with high-dimensional node features. By utilizing multiple classifiers to obtain multiple prediction results and fusing them through fuzzy integrals, it can improve overall classification performance and reduce errors caused by individual classifiers to a certain extent. Compared with mainstream classification methods in current graph neural network research, this invention improves classification results in terms of accuracy and recall, significantly enhancing the accuracy and stability of node classification, and providing an innovative and efficient solution for node classification tasks in graph networks. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further explained below with reference to specific embodiments, but this explanation does not limit the invention.
[0038] To improve the overall performance of citation node classification, this implementation plan starts from feature fusion at different levels and dimensions, considers the various effective semantic information existing in the network, and proposes a citation network node classification method that integrates GCN and multi-classifier. The specific steps are as follows:
[0039] S1: The knowledge citation structure based on academic fields regards the entire citation network as an undirected graph G that includes node attributes and topology. Each academic paper is regarded as a node in the network, and the citation relationship between documents is regarded as an edge between nodes. When a paper cites another paper, it means that an edge is established between the two papers.
[0040] By transforming the knowledge citation structure in academic fields into a network graph, the nodes in the network graph can be studied using graph theory and network analysis methods.
[0041] S2: Calculate the adjacency matrix and characteristic matrix X of the undirected graph G, where, N is the number of nodes, C is the number of features for each node, C features represent whether C words in the feature word dictionary appear in the citation, and the i-th row of the feature matrix X represents the feature vector of the i-th node.
[0042] S3: Initialize the edge weight matrix of the undirected graph G, and calculate the edge weights of the undirected graph G based on the feature matrix;
[0043] The formula for calculating the edge weight w(d1,d2) between any two nodes d1 and d2 in an undirected graph G is as follows:
[0044]
[0045] In the formula: F1 and F2 are the feature word sets of nodes d1 and d2, respectively; |F1∩F2| represents the size of the intersection of the feature word sets of nodes d1 and d2; and |F1∪F2| represents the size of the union of the feature word sets of nodes d1 and d2.
[0046] The edge weight between two documents depends on the ratio of the number of common feature words they contain to the total number of feature words C they contain. The more common feature words two documents have, the higher the edge weight between them, meaning the two nodes are more similar, and vice versa.
[0047] S4: Aggregate the edge weights calculated in S3 to the nodes at both ends of the edge, update the feature matrix X, and obtain the updated feature matrix;
[0048] Due to the characteristics of citation networks, when a feature in a node is zero, it indicates that the word does not appear in that paper node. Therefore, when updating node features based on edge weights, this feature should be skipped. The i-th row of the updated feature matrix X (i.e., the updated feature vector of the i-th node) is... ) is represented as:
[0049]
[0050] In the formula, N(i) represents the set of neighboring nodes of the i-th node, w(i,j) represents the edge weight between the i-th node and the j-th node, and X... j This represents the j-th row of the feature matrix X before the update (i.e., the feature vector of the j-th node);
[0051] S5: To further improve model performance, the features in the updated feature matrix obtained in S4 are filtered and amplified.
[0052] Filtering and amplification can be used to filter out inconspicuous features and amplify obvious features, thereby achieving the goals of removing noise, highlighting important features, and cleaning up low-correlation features.
[0053] The filtering and amplification steps are as follows:
[0054] First, calculate the variance of each feature;
[0055] Then, define a threshold θ var Features with variance greater than the threshold are retained.
[0056] Finally, it will be greater than the preset threshold θ f The eigenvalues are increased by a corresponding multiple based on the importance index to amplify the more prominent features.
[0057] As an improvement to the technical solution, S5 also includes the following steps: standardizing the amplified features so that they have a distribution with a mean of 0 and a standard deviation of 1.
[0058] After the above aggregation operation, the feature matrix X corresponding to the undirected graph G of the citation network can simultaneously contain the semantic information of the edges and neighboring nodes in the graph, and filter and amplify the attribute values in the feature matrix, thereby initially improving the performance of the model in the preprocessing stage.
[0059] S6: Input the adjacency matrix obtained in S2 and the feature matrix obtained in S5 into n classifiers respectively to obtain n first classification result matrices M corresponding one-to-one with the n classifiers, where n is greater than 1. The value in the i-th row and j-th column of the first classification result matrix M corresponding to each classifier represents the predicted value of the i-th node belonging to the j-th category.
[0060] The n classifiers use the same GCN model, and each GCN model is trained independently to obtain n trained GCN models. The n trained GCN models are obtained through different random initializations and randomness of the input data order.
[0061] S7: The n first classification result matrices M obtained in S6 are fused using fuzzy integral fusion to obtain the final classification result matrix Out, where, N is the number of nodes, Y is the number of categories, and the value in the i-th row and j-th column of the final classification result matrix Out represents the predicted value of the i-th node belonging to the j-th category;
[0062] Fuzzy integral is a method that integrates an input function with a fuzzy measure, and it can handle uncertainty in fuzzy environments. In this invention, the mathematical basis for fusion is the Sugeno fuzzy integral. The Sugeno fuzzy integral is a commonly used fuzzy integral method that calculates the output value based on the values of the fuzzy measure and the input function. It can effectively fuse multiple information sources, handle uncertainty and fuzziness, and is widely used in multi-classifier fusion and decision-making processes.
[0063] Preferably, fusing n first classification result matrices M through fuzzy integral fusion includes the following steps:
[0064] S71: For each category, extract the corresponding columns from the n first classification result matrices M corresponding to the n classifiers to form Y second classification result matrices Z that correspond one-to-one with each category, where... N is the number of nodes, and n represents the number of classifiers. The method to obtain the second classification result matrix Z(k) corresponding to the kth category is as follows: extract the kth column from the first classification result matrix M corresponding to each classifier to form the second classification result matrix Z(k) corresponding to the kth category.
[0065] S72: Taking the second classification result matrix Z corresponding to each category as input, the Sugeno fuzzy integral function is used to calculate the fusion result for each category. The formula for calculating the fusion result of the kth category is as follows:
[0066]
[0067] Where Fused(k) represents the fusion result of the k-th category, which is a vector; μ(i) is the fuzzy measure, representing the weight or confidence corresponding to the i-th classifier; and Z(k) is the fuzzy measure. i This represents the i-th column in the second classification result matrix Z(k) corresponding to the k-th category;
[0068] S73: Combine the fusion results of each category to obtain the final classification result matrix.
[0069] S8: Select the category corresponding to the largest value in each row from the final classification result matrix obtained from S7 as the classification result of the corresponding node, thus realizing the classification of the paper.
[0070] This invention also proposes a citation network node classification system that integrates GCN and multi-classifiers, for performing the aforementioned citation network node classification method that integrates GCN and multi-classifiers.
[0071] This invention presents a citation network node classification method (classification system) that integrates GCN and multi-classifiers. It re-integrates feature information from different levels and dimensions, fusing, filtering, and amplifying low-dimensional edge weights with high-dimensional node features. By utilizing multiple classifiers to obtain multiple prediction results and fusing them through fuzzy integrals, it improves overall classification performance and reduces errors caused by individual classifiers to some extent. Compared with mainstream classification methods in current graph neural network research, this invention improves classification results in terms of accuracy and recall, significantly enhancing the accuracy and stability of node classification. It provides an innovative and efficient solution for node classification tasks in graph networks.
[0072] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Python and the interpreted scripting language JavaScript.
[0073] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.
[0074] It should be understood that the present invention is not limited to the content already described above, and various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.
Claims
1. A citation network node classification method integrating GCN and multi-classifier, characterized in that, include: S1: The knowledge citation structure based on academic fields regards the entire citation network as an undirected graph G that includes node attributes and topology. Each academic paper is regarded as a node in the network, and the citation relationship between documents is regarded as the edge between nodes. S2: Calculate the adjacency matrix and characteristic matrix X of the undirected graph G, where, N is the number of nodes, C is the number of features for each node, and C features represent whether C words in the feature word dictionary appear in the citation. The i-th row of the feature matrix X represents the feature vector of the i-th node. S3: Initialize the edge weight matrix of the undirected graph G, and calculate the edge weights of the undirected graph G based on the feature matrix; S4: Aggregate the edge weights calculated in S3 to the nodes at both ends of the edge, and update the feature matrix X; S5: Filter and amplify the features in the updated feature matrix obtained in S4; S6: Input the adjacency matrix obtained in S2 and the feature matrix obtained in S5 into n classifiers respectively to obtain n first classification result matrices M corresponding one-to-one with the n classifiers, where n is greater than 1. The value in the i-th row and j-th column of the first classification result matrix M corresponding to each classifier represents the predicted value of the i-th node belonging to the j-th category. Among them, the n classifiers all use the same GCN model structure, and each GCN model is trained independently to obtain n trained GCN models; S7: The n first classification result matrices M obtained in S6 are fused using fuzzy integral fusion to obtain the final classification result matrix Out, where, N is the number of nodes, Y is the number of categories, and the value in the i-th row and j-th column of the final classification result matrix Out represents the predicted value of the i-th node belonging to the j-th category; S8: Select the category corresponding to the largest value in each row from the final classification result matrix obtained from S7 as the classification result of the corresponding node, thus realizing the classification of the paper.
2. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S3, the formula for calculating the edge weight w(d1,d2) between any two nodes d1 and d2 in the undirected graph G is: In the formula: F1 and F2 are the feature word sets of nodes d1 and d2, respectively; |F1∩F2| represents the size of the intersection of the feature word sets of nodes d1 and d2; and |F1∪F2| represents the size of the union of the feature word sets of nodes d1 and d2.
3. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S4, the feature matrix X is updated using the following formula: In the formula, Let X represent the i-th row of the updated feature matrix X, N(i) represent the set of neighboring nodes of the i-th node, and w(i,j) represent the edge weight between the i-th node and the j-th node. j This represents the j-th row of the feature matrix X before the update.
4. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S5, the filtering and amplification steps are as follows: First, calculate the variance of each feature; Then, define a threshold θ var Features with variance greater than the threshold are retained. Finally, it will be greater than the preset threshold θ f The eigenvalues are increased by a corresponding multiple based on the importance index to amplify the more prominent features.
5. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: S5 also includes a step of standardizing the amplified features to give them a distribution with a mean of 0 and a standard deviation of 1.
6. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S6, n trained GCN models are obtained through different random initializations and randomization of the input data order.
7. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S7, Sugeno fuzzy integrals are used for fusion.
8. The citation network node classification method integrating GCN and multi-classifier according to claim 1, characterized in that: In S7, fusing n first classification result matrices M through fuzzy integral fusion includes the following steps: S71: For each category, extract the corresponding columns from the n first classification result matrices M corresponding to the n classifiers to form Y second classification result matrices Z that correspond one-to-one with each category, where... N is the number of nodes, and n represents the number of classifiers. The method to obtain the second classification result matrix Z(k) corresponding to the kth category is as follows: extract the kth column from the first classification result matrix M corresponding to each classifier to form the second classification result matrix Z(k) corresponding to the kth category. S72: Taking the second classification result matrix Z corresponding to each category as input, the Sugeno fuzzy integral function is used to calculate the fusion result for each category. The formula for calculating the fusion result of the kth category is as follows: Where Fused(k) represents the fusion result of the k-th category, μ(i) is the fuzzy measure, representing the weight or confidence corresponding to the i-th classifier, and Z(k) i This represents the i-th column in the second classification result matrix Z(k) corresponding to the k-th category; S73: Combine the fusion results of each category to obtain the final classification result matrix.
9. A citation network node classification system integrating GCN and multi-classifier, characterized in that: The citation network node classification method for performing any one of claims 1-8, which combines GCN and a multi-classifier.