Diversifying Information Needs in Results of Question Retrieval

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1432–1436, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Diversifying Information Needs in Results of Question Retrieval Yaoyun Zhang, Xiaolong Wang, Xuan Wang, Ruifeng Xu, Jun Xu and Shixi Fan Key Laboratory of Network-oriented Intelligent Computation Harbin Institute of Technology, Shenzhen Graduate School, China {xiaoni5122, xuruifeng.hitsz, hit.xujun}@gmail.com {wangxl, wangxuan}@insun.hit.edu.cn Abstract Information need is an important factor in question retrieval. This paper proposes a method to diversify the results of question retrieval in term of types of information needs. CogQTaxo, a question hierarchy is leveraged to represent users’ information needs cognitively from three linguistic levels. Based on a prediction model of question types, three factors, i.e., scores of IR model, question type similarity and question type novelty are linearly combined to re-rank the retrieved questions. Preliminary experimental results show that the proposed method enhances the question retrieval performance in information coverage and diversity. 1 Introduction Most current question retrieval system attempts to fetch questions semantically similar to the query question (Jeon et al, 2005), together with the accepted answers from a large question-answer pair archive. Previous works focus on reducing the lexicon gap between the query question and retrieved questions (Cao et al, 2010) or recognize the single question type, i.e., the type of information needs(infoNeeds) in the query, and confine the types of retrieved questions to be the same as the query (Lytinen and Tomuro, 2002). Normally, the retrieved questions are ranked according to the semantic similarity to the query question. However, Taylor (1962) argues that the user may fail to express his infoNeeds fully in the question. Besides, given different contextual situations, users may have different intentions, which lead to different infoNeeds for the same question (Small and Strzalkowski, 2008). For an example question q1, “which bank provides the best credit card?”, if the user wants to confirm the bank he knows, the name of the bank is enough for an answer; while the user plans to open a credit card account, he may want to obtain detailed descriptions and comparisons between credit card services of different banks in addition to a single bank name. Furthermore, a play-it-safe user may expect the information source of the answer to be of authority or expertise, while a casual user may expect it to be commonsense that anyone can answer. Considering these requirement, the following two questions q2 and q3 should be provided to the user under a certain context. Nevertheless, such infoNeeds are not given explicitly in the q1. q2: Which bank should I choose for credit card, Citi Bank and Bank of America? q3: How to choose credit card? As can be observed, the three questions have different types, which are entity, alternative and method, respectively (Diekema et al, 2003). Apparently, the single-dimensional question taxonomies employed at present are insufficient to model those aspects of users’ infoNeeds (Pomerantz, 2005). Thus, more comprehensive question taxonomy is needed. The question retrieval results should also be diversified accordingly to …

Content truncated to 500 words