Natural Language Processing

  • Open Domain QA

  • We developed a question answering system - the Academia Sinica Question-Answering System (ASQA), which outputs exact answers for factoid questions of six types, i.e., person name, location name, organization name, artifact, time, and number. For example, when receiving a question like "誰是美國總統?", ASQA will search the document database and return the exact answer "布希". The architecture of ASQA comprises four main components, namely, Question Processing, Passage Retrieval, Answer Extraction, and Answer Ranking. Questions are analyzed to obtain question types, keywords, and focuses. Through a simple mapping table, question types are used to constrain possible answer types. Documents are segmented and indexed with both characters and words. After question analysis, queries are constructed from keywords to retrieve possible document passages, which are then sent to a named entity recognition system to obtain answer candidates. Finally, answers are ranked based on the needs of question focuses. Our proposed ASQA successfully combines machine learning and knowledge-based approaches to answer Chinese factoid questions, achieving approximately fifty accuracy for the Top 1 answers.
  • webiconDemo site :