Integrating linguistic patterns and term-entity associations in Chinese person description extraction

Person description extraction is an important task in biography generation, question answering and summarization. Most previous extraction approaches select descriptive passages depending on sentence structure and/or word co-occurrence information. In this paper, we focus on Chinese person description extraction verification by measuring the associations between the recognized person entities and the surrounding terms, called Term-Entity associations. The associations are derived from both the semantic knowledge provided in a Chinese well-known thesaurus HowNet and the term distributional information gathered from the news corpus. Relying on Term-Entity associations, the ineligible extracted descriptions could be filtered out so that the higher precision could be achieved in turn. As far as we know, no work on Chinese person description extraction has been reported in the literature.