TREC 2005 genomics track experiments at IBM watson

This paper describes our experiments in the TREC 2005 Genomics Track. For the ad-hoc retrieval task, we study synonym-based query expan-sion, as well as the effectiveness of a new pseudo-relevance feedback method which is derived from our recent work on semi-supervised learning. For t le categorization task, we study various methods for estimating conditional class probability and determining the optimal threshold parameter essential for obtaining high performance results for this task.