Context-based Quasi-Synonym Extraction

Nearor quasi-synonyms can have an important role in the task of Information Retrieval since they may help to address the problem of vocabulary mismatch between queries and documents. One current approach to generating quasi-synonyms uses a general similarity measure to score the synonymy of two words based on their context vectors. In contrast, we compare two simple measures that take into account more directly the contextual evidence that supports a synonymy relation. Experimental results using the Google n-gram collection show that our methods produce better synonyms than existing approaches.