Author: Noriko
Tomuro
Journal-ref: To appear in New Directions in Question Answering (AAAI
Press), 2004..
In this paper, we propose new directions in question-answering for retrieval models and learning with Frequently Asked Questions (FAQ) files. The idea of FAQ files has been around for some time, notably in Usenet FAQ files and call-center manuals. FAQ files became even more popular in recent years as the World Wide Web has become widely accessible. Therefore, developing application systems and techniques tailored to FAQ files is an important thrust in question-answering. By the nature of FAQ files, question-answering techniques involving FAQ files center on answer re-use, since these files contains pairs of previously asked questions and their answers. We describe on-going research on a realworld FAQ-based question-answering system called FAQFinder. We first describe the system’s current strategy for matching user questions against FAQ questions and report the system’s retrieval performance to date. Then we present our latest work on matching user questions against FAQ answers. Finally we discuss future research directions, including Q&A learning, that apply to FAQ-based question-answering systems in general.
Paper: Final Submission version (pdf 199k)