A Practical QA System In Restricted Domains

This paper describes an on-going research for a practical question answering system for a home agent robot. Because the main concern of the QA system for the home robot is the precision, rather than coverage (No answer is better than wrong answers), our approach is try to achieve high accuracy in QA. We restrict the question domains and extract answers from the pre-selected, semi-structured documents on the Internet. A named entity tagger and a dependency parser are used to analyze the question accurately. User profiling and inference rules are used to infer hidden information that is required for finding a precise answer. Testing with a small set of queries on weather domain, the QA system showed 90.9% of precision and 75.0% of recall.