Completely-Arbitrary Passage Retrieval in Language Modeling Approach

Abstract

Passage retrieval has been expected to be an alternative method to resolve length-normalization problem, since passages have more uniform lengths and topics, than documents. An important issue in the passage retrieval is to determine the type of the passage. Among several different passage types, the arbitrary passage type which dynamically varies according to query has shown the best performance. However, the previous arbitrary passage type is not fully examined, since it still uses the fixed-length restriction such as n consequent words. This paper proposes a new type of passage, namely completely-arbitrary passages by eliminating all possible restrictions of passage on both lengths and starting positions, and by extremely relaxing the type of the original arbitrary passage. The main advantage using completely-arbitrary passages is that the proximity feature of query terms can be well-supported in the passage retrieval, while the non-completely arbitrary passage cannot clearly support. Experimental result extensively shows that the passage retrieval using the completely-arbitrary passage significantly improves the document retrieval, as well as the passage retrieval using previous non-completely arbitrary passages, on six standard TREC test collections, in the context of language modeling approaches.