Lightly-Supervised Attribute Extraction

Web search engines can greatly benefit from knowledge about a ttributes of entities present in search queries. In this paper, we introduce light ly-supervised methods for extracting entity attributes from natural language tex t. Using these methods, we are able to extract large numbers of attributes of differe nt entities at fairly high precision from a large natural language corpus. We compare our methods against a previously proposed pattern-based relation extractor, s howing that the new meth- ods give considerable improvements over that baseline. We also demonstrate that query expansion using extracted attributes improves retrieval performance on un- derspecified information-seeking queries.