NTCIR-5 Query Expansion Experiments using Term Dependence Models
Koji Eguchi · 2005
This paper reports the results of our experiments performed for the Query Term Expansion Subtask, a subtask of the WEB Task, at the Fifth NTCIR Workshop, and the results of our further experiments. In this paper we mainly investigated: (i) the effectiveness of query formulation by composing or decomposing compound words and phrases of the Japanese language, which is based on a theoretical framework via Markov random fields, but taking into account special features of the Japanese language; and (ii) the effectiveness of the combination of phrase-based query formulation and pseudo-relevance feedback. We showed that pseudo-relevance feedback worked well, particularly when using query formulation with compound words.