Query Formulation by Selecting Good Terms

Query Formulation by Selecting Good Terms 李佳蓉 Chia-Jung Lee, 林怡君 Yi-Chun Lin, 陳瑞呈 Ruey-Cheng Chen 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University {cjlee1010, yi.crystal, rueycheng}@gmail.com 劉培森 Pei-Sen Liu 資訊工業策進會 Institute for Information Industry psliu@iii.org.tw 鄭卜壬 Pu-Jen Cheng 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University pjcheng@csie.ntu.edu.tw Abstract It is difficult for users to formulate appropriate queries for search. In this paper, we propose an approach to query term selection by measuring the effectiveness of a query term in IR systems based on its linguistic and statistical properties in document collections. Two query formulation algorithms are presented for improving IR performance. Experiments on NTCIR-4 and NTCIR-5 ad-hoc IR tasks demonstrate that the algorithms can significantly improve the retrieval performance by 9.2% averagely, compared to the performance of the original queries given in the benchmarks. Experiments also show that our method can be applied to query expansion and works satisfactorily in selection of good expansion terms. Keywords: Query Formulation, Query Term Selection, Query Expansion. 1. Introduction Users are often supposed to give effective queries so that the return of an information retrieval (IR) system is anticipated to cater to their information needs. One major challenge they face is what terms should be generated when formulating the queries. The general assumption of previous work [14] is that nouns or noun phrases are more informative than other parts of speech (POS), and longer queries could provide more information about the underlying information need. However, are the query terms that the users believe to be well-performing really effective in IR? Consider the following description of the information need of a user, which is an example description query in NTCIR-4: Find articles containing the reasons for NBA Star Michael Jordan's retirement and what effect it had on the Chicago Bulls. Removing stop words is a common way to form a query such as “contain, reason, NBA Star, Michael Jordan, retirement, effect, had, Chicago Bulls”, which scores a mean average precision (MAP) of 0.1914. It appears obviously that terms contain and had carry relatively less information about the topic. Thus, we take merely nouns into account and generate another query, “reason, NBA Star, 69 Michael Jordan, retirement, effect, Chicago Bulls”, which achieves a better MAP of 0.2095. When carefully analyzing these terms, one could find that the meaning of Michael Jordan is more precise than that of NBA Star, and hence we improve MAP by 14% by removing NBA Star. Yet interestingly, the performance of removing Michael Jordan is not as worse as we think it would be. This might be resulted from that Michael Jordan is a famous NBA Star in Chicago Bulls. However, what if other terms such as reason and effect are excluded? There is no explicit clue to help users determine what terms are effective in an IR system, especially when they lack experience of searching documents in a specific domain. Without comprehensively understanding the document collection to be retrieved, it is difficult for users to …

Content truncated to 500 words