An Introduction to BLCU Personal Attributes Extraction System

摘要

We describe our methods for share task of personal attributes extraction. We divide all 25 attributes into several categories and propose 4 kinds of pipelines to carry out value extraction. There are two stages in the process. The first stage uses CRF model or regular expression based extractor to produce initial answers. In the second stage, we propose two methods to filter out mistake answers: protagonist dependency relationship based filter and attribute keywords based filter.