A Re-ranking Algorithm Based on Focused Named Entities

This paper proposed a new method for learning to re-rank the retrieved documents based on the evaluation of the semantic relevance between named-entities in these documents and the query words, especially relevance between the query and the most topical named entities in these documents. The relevance weights used to rank documents were evaluated by analyzing the co-occurrence characters of focused named entities with respect to query. In this method, firstly, given the set of retrieved documents containing a query, the focused named entities in these documents are recognized; secondly, the relevance level of the query with respect to the focused entities in each retrieved document is estimated; thirdly, these retrieved documents are re-ranked with these relevance levels. Moreover, Experimental results on SEWM2006 test set indicate that our method can work well.