Hypergraph-based Document Representation and Hyperedge Similarity Algorithm

To address the limitations of traditional document representation models in capturing complex semantic relationships, this paper proposes a novel hypergraph-based document representation model. In this model, words are represented as nodes, while documents are represented as hyperedges, resulting in a hypergraph structure that reveals intricate multi-dimensional relationships and structural information within documents. Hypergraphs were constructed using four datasets, which include Chinese and English news and poetry. An analysis of network topology and hyperdegree distribution revealed small-world and scale-free properties. To overcome the limitation of existing hyperedge similarity algorithms, which treat all nodes as equivalent, we developed the Node-Weighted Cosine-Hamming Hyperedge Similarity (NWCHHES) algorithm. Classification experiments indicated that NWCHHES outperformed existing methods across multiple metrics, providing more accurate hyperedge similarity calculations.