Scholarlink

<p>With the continuous development of graph networks and deep learning, the idea of constructing deep learning models on graphs shows great potential, in which various models with graph convolutional neural network (GCN) as the core play an important role in the node classification task. GCN can aggregate semantic information and topology in the network to realize the node classification results, but it can also fail to comprehensively cover the effective semantic information due to the complexity of the network, which resulting in unsatisfactory classification results. Therefore, this paper proposes an approach based on multiple classifiers fusion of GCN models (Multiple Classifiers Fusion Graph Convolutional Neural Network, MCFGCN). Firstly, this method fuses and filters the feature information during the aggregation process in order to consider the effective semantic information in the graph network more comprehensively. Then, multiple GCN models are initialized to train the features. And finally, obtains the prediction results of multiple models, fuses them using fuzzy integrals and obtains the final classification results of the nodes, so as to obtain a more accurate and stable classification result. In order to verify the superiority of the proposed method, MCFGCN is compared with the mainstream algorithms of graph neural network research, GCN, GAT, GraphSAGE, and two other traditional classification methods, MLP and Node2Vec, on three citation network mainstream datasets. The experimental results show that MCFCCN improved accuracy, recall, and F1 score by 2.8%, 2.1%, and 3.8% on the Cora dataset, respectively, compared to the baseline GCN model. On the Citeseer dataset, the accuracy has improved by 1% compared to the advanced model (GraphSAGE). On the Pubmed dataset, MCFCCN improved accuracy by 1.1% compared to GAT. Overall, MCFCCN highlights its robustness and effectiveness in multiple citation network datasets through the combination of multi classifier fusion and fuzzy integration, improving the effectiveness of node classification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improving node classification method using GCN and fused multi-classifiers

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Xiaoli Bo,
  • Yang Yu

摘要

With the continuous development of graph networks and deep learning, the idea of constructing deep learning models on graphs shows great potential, in which various models with graph convolutional neural network (GCN) as the core play an important role in the node classification task. GCN can aggregate semantic information and topology in the network to realize the node classification results, but it can also fail to comprehensively cover the effective semantic information due to the complexity of the network, which resulting in unsatisfactory classification results. Therefore, this paper proposes an approach based on multiple classifiers fusion of GCN models (Multiple Classifiers Fusion Graph Convolutional Neural Network, MCFGCN). Firstly, this method fuses and filters the feature information during the aggregation process in order to consider the effective semantic information in the graph network more comprehensively. Then, multiple GCN models are initialized to train the features. And finally, obtains the prediction results of multiple models, fuses them using fuzzy integrals and obtains the final classification results of the nodes, so as to obtain a more accurate and stable classification result. In order to verify the superiority of the proposed method, MCFGCN is compared with the mainstream algorithms of graph neural network research, GCN, GAT, GraphSAGE, and two other traditional classification methods, MLP and Node2Vec, on three citation network mainstream datasets. The experimental results show that MCFCCN improved accuracy, recall, and F1 score by 2.8%, 2.1%, and 3.8% on the Cora dataset, respectively, compared to the baseline GCN model. On the Citeseer dataset, the accuracy has improved by 1% compared to the advanced model (GraphSAGE). On the Pubmed dataset, MCFCCN improved accuracy by 1.1% compared to GAT. Overall, MCFCCN highlights its robustness and effectiveness in multiple citation network datasets through the combination of multi classifier fusion and fuzzy integration, improving the effectiveness of node classification.