Seminario Interdipartimentale di Algoritmica  

Monday, December 19, 2005, 12:00 noon
Ranking Word Senses with the HITS Algorithm: Word Sense Disambiguation Meets Link Analysis
Roberto Navigli, Università di Roma "La Sapienza"

DIS - Department of Computer and System Sciences
Room C3, second floor

Abstract:

The language we use in everyday life is ambiguous, as we can express different concepts with the same word. Word Sense Disambiguation (WSD) concerns the automatic assignment of appropriate senses to a set of words in context. Senses can be chosen from computational lexicons, i.e. machine-readable dictionaries encoding the possible interpretations of each word (e.g. bank as a financial institution and as a steep slope beside a body of water). Based on graph theory, link analysis is the process of building up networks of interconnected entities in order to explore patterns and trends. In this talk, we will briefly introduce both areas and will then focus on the application of link analysis to WSD. In knowledge-based WSD, we can represent a computational lexicon as a graph G = (V,E) where V is the set of word senses and E is the set of semantic relations holding betwen pairs of senses. Specifically, we will introduce a knowledge-based algorithm, called Structural Semantic Interconnections, and describe its implementation in terms of an adaptation of Kleinberg's HITS algorithm to sense relevance ranking. Finally, we will provide an evaluation of the algorithm, and explore some open problems.