We built a domain-specific Chinese interactive Question An-swering(QA) system which has already been available as a public service via phone text messaging. The system utilizes Topic Forest as the dialog management model to be capable of keeping track of users' interests. The Question Answering component is a hybrid approach which consists of both a community Question Answering engine and a new knowledge-based QA engine. In the new QA engine, we constructed a semantic pattern matching model to automatically translate question topics and targets generated by the natural language understanding unit to SPARQL queries, which eventually build the final answer. Through experimental data collected from real user survey and case study, our system shows promising results with the comparison over the practical QA system Siri as the industrial standard in terms of both accuracy and user satisfaction rates.
展开▼