Verifying person descriptions with term-entity association

Person description extraction is an important task in biography generation, question answering and summarization, etc. While most of the previous extraction methods mainly depended on structural information, the work presented in the paper focuses on extraction verification by integrating linguistic knowledge provided by HowNet (with semantic knowledge) and the newswire corpus (with statistical information), from which the associations between terms (i.e. the words in HowNet) and person entities are measured. With term-entity association, ineligible descriptions extracted could be filtered out, and a higher precision is achieved in turn.