Natural Language Processing

  • FAQ-Centered QA

  • The FAQ-Centered QA system we created attempts to answer questions about small domains, such as an organization like Academia Sinica. The knowledge needed for such a small domain is controllable and FAQ lists usually exist. Therefore, very little effort is needed to collect FAQ pairs. The main goal of an FAQ QA system is to match an input question with the most appropriate FAQ, then output the corresponding answers. In our system, we try to parameterize the FAQ list. After generalization, some FAQs can be categorized into the same parameterized FAQ. This process has two major benefits. First, it gives semantic analysts an overview of the FAQs and helps generate our tree-like knowledge base, InfoMap. Also, since every question matches a parameterized FAQ, it could be used to generate related information, not just the answer the user wants. We use InfoMap to organize the knowledge for answering FAQs. Three kinds of knowledge, namely, concepts, relations, and syntactic templates, are stored in InfoMap. After receiving a question, a mechanism fires the templates and the related concepts through the relations between them. A scoring method is then applied to rank the matched FAQs.
  • Demo Site URL: