Language Models for Genomics Information Retrieval: UIUC at TREC 2007 Genomics Track

The University of Illinois at Urbana-Champaign (UIUC) participated in TREC 2007 Genomics Track. Our gen- eral goal of participation is to apply language model- based approaches to the genomics retrieval task and study how we may extend the standard language models to ac- commodate two special needs for this year's genomics retrieval task: (1) gene synonym expansion and (2) con- junctive query interpretation. We also tried user rele- vance feedback. Preliminary result analysis shows that our synonym expansion method can improve document- level MAP, but generally has little influence on passage- level and aspect measures, while conjunctive scoring is not as effective as the standard KL-Divergence scoring, even though our pre-TREC experiments on a small set of training data showed otherwise. Relevance feedback appears to help. Further experiments and analysis are needed to draw more definitive conclusions.