York University at TREC 2006: Genomics Track

Our Genomics experiments mainly focus on addressing four problems in biomedical information retrieval. The four problems are: (1) how to deal with synonyms? (2) how to deal with the frequent use of acronyms? (3) how to deal with homonyms? (4) how to deal with the document-level retrieval, passagelevel retrieval and aspect-level retrieval? In particular, we use the automatic query expansion algorithm proposed at TREC 2005 to construct structured queries for document-level retrieval and we also apply several data mining techniques for passage-level retrieval and aspect-level retrieval. The mean average precisions (MAP) for our automatic run “york06ga1” are 0.3365 at the document-level retrieval, 0.0197 at the passage-level retrieval and 0.1084 at the aspect-level retrieval. The evaluation results show that the automatic query expansion algorithm is effective for improving document-level retrieval performance. However, our retrieval performance on passage-level and aspect-level is poor. One possible reason is that we did not follow the TREC 2006 Genomics track protocol regarding the calculation of passage measures correctly. Therefore, we built a completely wrong index for the passage-level retrieval. Since our aspectlevel retrieval is based on the results obtained from our passage level retrieval, the results thus obtained are sub-optimal.