Book structure and genre classification using GNNs

Graph Attention Networks (GATs) are a specific form of Graph Neural Networks (GNNs) that aggregate hidden features from nodes and use self-attention mechanisms to re-weight edge values. This enables GATs to effectively discern the significance of various relationships within the graph. In this study, we leveraged these characteristics of GATs to convert text-based books into graph networks, train the model, and classify the genre of the books. Analyzing the performance of GATs, we explored a novel approach that converts text sentences into graph notation. By transforming text-based books into graph structures, we leveraged the strengths of GATs to classify the genres of the books. These datasets comprised book graphs, with entities represented as nodes and sentiment analysis scores of the relationships between these entities represented as edges. The datasets varied in graph size and construction, and we applied the same GAT model configurations across all tests to measure accuracy. The experimental results demonstrated the potential of GATs for successful book genre classification and provided valuable insights into how the size of input graphs affects GAT performance.