Entity-Based Relevance Feedback for Genomic List Answer Retrieval

In this paper we present a system which uses ontological resources and a gene name variation generation tool to expand concepts in the original query. The novelty of our approach lies in our concept-based normalization ranking model. For the 2007 Genomic task, we also modified this system architecture with an additional dynamic form of query expansion called entity-based relevance feedback. This technique works by identifying potentially relevant entity instances in an initial set of retrieved candidate paragraphs. These entities are added to the initial query with the aim of boasting the rank of passages containing lists of these entities. Our final modification to the system, aims to maximizing the passage-level MAP score, by dropping sentences that do not contain any query concepts, from the beginning and the end of a candidate paragraph. Our TREC 2007 results show that our relevance feedback method can significantly improve baseline retrieval performance with respect to docume...

• Type: Conference
• Year: 2007
• Where: TREC
• Authors: Nicola Stokes, Yi Li, Lawrence Cavedon, Eric Huang, Jiawen Rong, Justin Zobel
• Keywords: Candidate Paragraph, Initial Query, Relevance Feedback, TREC 2007, TREC 2008