The Mining of Term Semantic Relationships and its Application in Text Classification

This paper proposes an approach for mining the semantic relationships between terms. Using a dependency model based on syntactic parsing, the syntactic features of a term are first extracted from large scale corpus, and then the vector representation for this term is constructed. By the cosine similarities between vectors, we can get the semantically related words for a term. We apply the semantic knowledge to document vector representation in text classification. The experiment on the standard data sets shows that our approach gets a better performance compared with the traditional classifiers.