Semantic Textual Similarity Using Maximal Weighted Bipartite Graph Matching

The paper focused on developing a system to assess semantic similarity between sentences by employing maximal weighted bipartite graph matching of sentence tokens. It utilized two token similarity measures: WordNet-based similarity and a statistical word similarity measure to address WordNet's limitations. The study explored three systems: a simple bag of words tokenization, a detailed tokenization capturing named entities and other specific elements, and a system incorporating grammatical dependencies to capture token context.