Rutgers Information Interaction Lab at TREC 2005: Trying HARD

Within the structure of the TREC 2005 HARD track guidelines, we investigated the following hypotheses: H1: Query expansion using a “clarity”-based approach will increase effectiveness over baseline queries and baseline queries plus pseudo-relevance feedback; H2: Query expansion based on the Web will increase effectiveness over baseline queries and baseline queries plus pseudo-relevance feedback; H3: Query expansion using terms selected by the searcher from those suggested by clarity modeling and/or the web will increase effectiveness over baseline queries, baseline queries plus pseudo-relevance feedback, and queries expanded by all suggested terms; H4: Query expansion using “problem statements” elicited from the searcher will increase effectiveness over baseline queries and baseline queries plus pseudo-relevance feedback; H5: The effectiveness of query expansion using problem statements will be negatively correlated with query “clarity”. H1 and H2 were tested without user intervention; H3 and H4 were tested using two different “clarification forms”; H5 was tested using the results of the H4 clarification form. Baseline queries were generated from the topic titles and descriptions; query expansion was accomplished by adding terms to the baseline queries, with a variety of weights given to the expansion terms, relative to the baseline terms. Preliminary results indicate that H1, H2, H3 and H4 are in part weakly supported, in that performance is increased over baseline, it is not increased over pseudo-relevance feedback. H5 was not supported. Combining some degree of user interaction (H3) with pseudo-relevance feedback appears to lead to increased performance.