ELECTRA Workshop on Methodologies and Evaluation of Lexical Cohesion Techniques in Real-world Applications (Beyond Bag of Words)

We compare different query formulation strategies and expansion based on lexical affinities in the context of passage retrieval. Our method to expand the queries using lexical affinities replaces only the missing terms from the original query in candidate passages while scoring them. The replacement term’s affinity with the missing term is used to weight the substitution, and the degree of affinity is computed using statistics generated from a terabyte corpus. The passages extracted using this replacement method and a set of passages extracted using different formulation strategies are evaluated using TREC’s QA test set.