Answering Definition Questions Using Web Knowledge Bases

R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 498 – 506, 2005. © Springer-Verlag Berlin Heidelberg 2005 Answering Definition Questions Using Web Knowledge Bases Zhushuo Zhang, Yaqian Zhou, Xuanjing Huang, and Lide Wu Department of Computer Science and Engineering, Fudan University, Shanghai, China, 200433 {zs_zhang, zhouyaqian, xjhuang, ldwu}@fudan.edu.cn Abstract. This paper presents a definition question answering approach, which is capable of mining textual definitions from large collections of documents. In order to automatically identify definition sentences from a large collection of documents, we utilize the existing definitions in the Web knowledge bases in- stead of hand-crafted rules or annotated corpus. Effective methods are adopted to make full use of Web knowledge bases, and they promise high quality response to definition questions. We applied our system in the TREC 2004 definition question-answering task and achieved an encouraging performance with the F- measure score of 0.404, which was ranked second among all the submitted runs. 1 Introduction When people want to learn an unknown concept from a large collection of documents, the most commonly used tools are the search engines. They submit a query to a search engine system, and the search engine returns a number of pages related to the query terms. Usually, the pages returned are ranked mainly based on keywords matching rather than their relevance to the query terms. The users have to read a lot of returned pages to organize the information they wanted by themselves. This procedure is time- consuming, and the information acquired is not concentrative. The research of Ques- tion Answering (QA) intends to resolve this problem by answering user’s questions with exact answers. Questions like “Who is Colin Powell?” or “What is mold?” are definition questions [3]. Their relatively frequent occurrences in logs of Web search engines [2] indicate that they are an important type of question. The Text REtrieval Conference (TREC) provides an entire evaluation for definition question answering from TREC2003. A typical definition QA system extracts definition nuggets that contain the most descrip- tive information about the question target (the concept for which information is being sought is called the target term, or simply, the target) from multiple documents. Until recently, definition questions remained a largely unexplored area of question answering. Standard factoid question answering technology, designed to extract single answers, cannot be directly applied to this task. The solution to this interesting re- search challenge will involve the techniques in related fields such as information extraction, multi-document summarization, and answer fusion. In order to extract definitional nuggets/sentences, most systems use various pattern matching approaches. Kouylekov et al. [10] relied on a set of hand-crafted rules to Answering Definition Questions Using Web Knowledge Bases 499 find definitional sentences. Sasha et al. [12] proposed to combine data-driven statisti- cal method and machine learned rules to generate definitions. Cui et al. [7] used soft patterns, which were generated by unsupervised learning. Such methods require hu- man labor to construct patterns or to annotate corpus more or less. Prager et al. [8] try …

Content truncated to 500 words