A Multi-Granularity Heterogeneous Graph for Extractive Text Summarization

Extractive text summarization selects the most important sentences from a document, preserves their original meaning, and produces an objective fact-based summary. It is faster less computationally intensive than abstract techniques. Learning cross-sentence relationships crucial for extractive summarization. However, of language models currently in use process data sequentially, which makes it difficult to capture such inter-sentence relations, especially long documents. This paper proposes model based on graph neural network (GNN) address this problem. The effectively represents using graph-structured document representation. In addition sentence nodes, we introduce two nodes with different granularity structure, words topics, bring levels semantic information. node representations are updated by attention (GAT). final summary obtained binary classification nodes. Our method was demonstrated be highly effective, as supported results our experiments CNN/DM NYT datasets. To specific, approach outperformed baseline same type terms ROUGE scores both datasets, indicating potential proposed enhancing tasks.