Utilizing semantic, syntactic, and question category information for automated digital reference services

Abstract

Digital reference services normally rely on human experts to provide quality answers to the user requests via online communication tools. As the services gain more popularity, more experts are needed to keep up with a growing demand. Alternatively, automated question answering module can help shorten the question-answering cycle. When the system receives a new user submitted question, the similarity of the user's request and the existing questions in the archive can be compared. If the appropriate match is found, the system then uses the associated answer to response to such request. Since a question is relatively short and two questions might contain very few words in common, the challenge is how to effectively identify the similarity of questions. In this paper, we focus on the problem of identifying questions that convey the similar information need. That is, our goal is to find paraphrases of the original questions. To achieve this, we propose a hybrid approach that combines semantic, syntactic, and question category to judge question similarity. Semantic and syntactic information is measured by taking into account word similarity, word order, and part of speech information. Information about the types of question is derived from a Support Vector Machine classifier. The experimental results demonstrate that our combined measures are highly effective in distinguishing original questions and their paraphrases, thus improving the potency of question matching task. © 2008 Springer Berlin Heidelberg.