International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 06 | Jun 2024
www.irjet.net
p-ISSN: 2395-0072
Benchmarking Traditional and Graph Neural Network Models for Node Classification in Literature Characters Kaina Shaikh1 1Department of Artificial Intelligence and Data Science, Vidya Pratishthan’s Kamalnayan Bajaj
Institute of Engineering & Technology, Baramati, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Node classification is a pivotal task in network
This paper presents a comparative analysis of various GNN models for node classification, specifically examining the interaction network of book characters in a novel. Each character is represented as a node, and interactions between characters are depicted as edges. We utilized three popular GNN architectures: Graph Convolutional Network (GCN), Graph Attention Network (GAT), and GraphSAGE. Additionally, we developed a fusion model that integrates these three GNNs to enhance classification performance.
analysis, focused on predicting node labels by leveraging both node features and the graph structure. Our study evaluates the effectiveness of Graph Neural Networks (GNNs) for node classification, concentrating on three models: Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), and GraphSAGE. We also propose a fusion model that integrates the strengths of GCN, GAT, and GraphSAGE to enhance classification accuracy. Experiments were conducted on a network derived from a book dataset, aiming to classify main versus supporting characters in a novel. We benchmarked these GNN models against traditional machine learning algorithms, including Random Forest and Support Vector Machine (SVM). Our results show that the fusion GNN model achieved the highest accuracy, surpassing individual GNN models and performing on par with traditional machine learning models. The performance of the fusion model highlights the potential of hybrid approaches in node classification tasks. This research provides valuable insights into the comparative advantages of different GNN architectures and their practical applications in network analysis.
To provide a comprehensive comparison, traditional machine learning models, including Random Forest and Support Vector Machine (SVM), were also implemented. The performance of each model was evaluated based on accuracy metrics, with the fusion GNN model achieving the highest average accuracy of 99.86%. This performance underscores the potential of GNNs, especially the fusion approach, in effectively handling node classification tasks in complex network data. By analyzing the interaction network of book characters, this study demonstrates the strengths and weaknesses of different GNN models and traditional machine learning approaches. The results provide valuable insights for researchers and practitioners working on similar node classification problems in various domains, especially in literary analysis and social network exploration within fictional settings.
Key Words: Node Classification, Network analysis, Graph Neural Networks (GNNs), Graph Convolutional Network (GCN), Graph Attention Network (GAT), GraphSAGE.
1.INTRODUCTION
2.RELATED WORK
One of the core tasks in graph-based learning is node classification, which involves making predictions about the labels of nodes in a network based on their attributes and relationships. Graph Neural Networks (GNNs) have emerged as effective tools for this purpose, exploiting graph structure to boost prediction accuracy.
The technique used in [1] for node classification is based on Graph Neural Networks (GNNs), notably Graph Convolutional Networks (GCN) and Graph Attention Networks. GNNs, particularly GCNs and GATs, have been acknowledged for their capacity to successfully capture structural information from complicated graphs. GCNs combine information from nearby nodes with properties from the core node, capturing both local and global relationships. In contrast, GATs use attention techniques to allow nodes to choose attend to informative neighbors throughout the aggregation process[3].
In this study, we focus on the relationships between characters of novel in a book dataset, where characters are represented as nodes and interactions between them are shown by edges. Understanding these relationships is critical for character analysis, narrative comprehension, and other literary studies. Our goal is to enhance the performance of node classification tasks associated with character interactions by identifying complex patterns through the application of sophisticated GNN models.
© 2024, IRJET
|
Impact Factor value: 8.226
The Graph Convolutional Network (GCN) model [2], which uses an effective layer-wise propagation rule developed from a first-order approximation of spectral convolutions on graphs, is the foundation for node categorization. The model
|
ISO 9001:2008 Certified Journal
| Page 1268