Article Preview
TopIntroduction
Question Reformulation is one of the components of Question Analysis module in Question Answering System. Question Reformulation reformulates the input question according to user’s need in order to affect the accuracy of subsequent modules. Why-type non-factoid questions are complex and ambiguous; making them difficult to answer. It is difficult to understand the actual need of user and derive an appropriate non-ambiguous meaning to it. If a correct query is posed to a search engine, it retrieves appropriate web pages that ultimately help in accurate document retrieval. In English language, there are two broad categorizations of questions (1) Factoid questions of type what, where, which, when and who; (2) Non-Factoid questions of type why and how. The factoid questions are simple and non-ambiguous whereas non-factoid questions are complex and difficult to answer.
Question Reformulation plays a crucial role in question answering system. It retransforms question into an appropriate query that depicts the user’s need and thus helps in efficient answer retrieval. The performance of question reformulation affects the performance of subsequent modules, i.e. document, answer candidate extraction and answer re-ranker (Kangavari et al., 2008).
Query reformulation is a key task in today’s web search engines for retrieving accurate and best results corresponding to the users’ query. Query reformulation is a process of modifying original query to resolve problems of ambiguity, vocabulary mismatch and vagueness. There are different techniques to query reformulation viz. (1) query expansion, (2) query suggestion and (3) query refinement (Ooi et al., 2015).
Query expansion expands query based on (a) relevance feedback by finding co-occurring terms, (b) query terms appended by their synonyms retrieved from WordNet and (c) retrieved informative terms for expansion from definition clusters (Bernhard, 2010). Query refinement modifies query based on the users’ past query logs. It doesn’t provide choice to user in selecting terms which can be appended to query. Terms are generated based on user feedback from the top ranked documents irrelevant to its appropriateness which helps in achieving high recall and precision. Finally query suggestion helps understanding the actual information need of user and is found as the most fundamental features of search engines. They are often required in case of rare query being posed, single-term query, unambiguous query suggestions, query suggestions are generalized form of original query and several pages are crawled by user. The approach suggests several other refined query corresponding to original user query based on the users’ interest/search logs analysis so that user can select terms that should be replaced original terms for better document retrieval.
The paper focuses on improving the why-question answering system by reformulating why-questions into an appropriate query that can depict the user’s need and when posed on search engine, help in retrieving appropriate web pages. There are some cases where the actual user need can’t be understood from the question, thus there comes the need for analyzing the question and reformulates it into an appropriate form that can depict the user need from the question.
The organization of paper is described as: Section 2 discusses researches of reformulation. Section 3 puts light on the main focus of the article. Section 4 analyzes different components of question with their impact on reformulation. Section 5 discusses algorithm designed for reformulation of different why-type questions. Section 6 highlights implementation details utilized while designing a user interface for reformulation. Section 7 describes results with their analysis on user feedback. Finally section 8 concludes the work.