Entity Profile Based Approach in Automatic Knowledge Finding.

摘要

We focus on the problem of profile building in this year s KBA track and proposed two methods. The first method is a baseline, which selects the stream items that has exact string match with the query entity. All the matched documents are assigned with the same relevance score. In the second method, we propose to use the related entities to help us identify the information related to the query entity. In particular, we retrieve the wikipedia pages for the query entities and extract the anchor text in all the internal links within the wikipedia page. These anchor text are treated as the related entities of the query entity and they are used to build the profile of the query entity. Given a stream item (i.e., a document), the relevance score is estimated by integrating the match with the query entity and the match with the related entities. Results on the training data show that the second method is more effective.