Using topic-concept based Clustering for Genomic Information Retrieval in TREC

In this paper, a topic-concepts Clustering method for the task of Genomic Information Retrieval in TREC is presented. The main idea is that all the documents are clustered according to the 36 queries, then, for each query, only the documents of the same cluster are considered as related documents and used to generate the accurate answer. It converts the queries and documents to concepts; then, expands the queries by using the pseudo-relevance feedback technique, and the expanded queries are considered initial centers to cluster all the documents; finally, the documents in each cluster are re-ranked respectively to mine the result for the corresponding query. The results of comparative experiments on the TREC 2007 Genomics indicate that this method is effective.