Integrating techniques of social network analysis and word embedding for word sense disambiguation
Chihli Hung, Chih-Neng Hung, Hsien-Ming Chou · Kybernetes · 2025
Purpose This research addresses the challenge of polysemous words in word embedding techniques, which are commonly used in text mining. It aims to resolve word sense ambiguity by introducing a social network sense disambiguation (SNSD) model based on social network analysis (SNA). Design/methodology/approach The SNSD model treats words as members of a social network and their co-occurrence relationships as interactions. By analyzing these interactions, the model identifies words with high betweenness centrality, which may act as bridges between different word sense communities, indicating polysemy. This unsupervised method does not rely on pre-tagged resources and is validated using the IMDb dataset. Findings The SNSD model effectively resolves word sense ambiguity in word embeddings, proving to be a cost-effective and adaptable solution to this issue. The experimental results demonstrate that the model enhances the accuracy of word embeddings by accurately identifying the correct meanings of polysemous words. Originality/value This study is the first to apply SNA to word sense disambiguation (WSD). The SNSD model offers a novel, unsupervised approach that overcomes the limitations of traditional supervised or knowledge-based methods, providing a valuable contribution to the field of text mining.