A Domain Independent Semantic Measure for Keyword Sense Disambiguation

Understanding the user's intention is crucial for many tasks that involve human-machine interaction. To that end, word sense disambiguation (WSD) techniques play an important role. WSD techniques typically require well-formed sentences as context to operate, as well as pre-defined catalogues of word senses. However, there are some scenarios on the Web where such conditions do not apply well, such as when there is a need to disambiguate keywords from a query, or sets of tags describing any Web resource, where the context does not come as well-formed sentences. In this paper, we propose an approach to disambiguate sets of keywords by linking them to concepts of a given ontology that is not known at training time. Our approach grounds on a semantic relatedness measure between words and concepts, and explores different disambiguation algorithms to study the contribution of both word and sentence-level representations. We focus on situations where the available linguistic information is very scarce (e.g., keyword-based / Web search queries), hampering natural language based approaches. Experimental results show the feasibility of our approach in general and in specific knowledge domains without previous training for the target domain.