Meta search models for online forum thread retrieval: research in progress

Abstract

Online forum thread retrieval is the task of retrieving threads that satisfy a user information need. Several thread representations have been proposed, and it has been found that combining these representations outperformed the retrieval using the individual representations. However, these combining methods leverage query relevance judgments to rank threads. Furthermore, in online forums, obtaining relevance judgments is not an option. As a result, in this paper, we propose to combine various thread representations using meta search techniques; many meta search techniques do not require training and has been found to produce a competitive result to the approaches that use relevance judgments. Our experimental result shows two things. First, combining thread representations using meta search techniques is an effective approach. Second, the CombSUM or the CombMNZ meta search techniques outperformed the best baseline method on high precision searches.