York University at TREC 2004: HARD and Genomics Tracks

York University participated in HARD and Genomics tracks this year. For both tracks, we used Okapi BSS (basic search system) as the basis. Our experiments mainly focused on exploiting various methods for combining document and passage scores, new term weighting formulae and feedback methods for query expansion. For HARD track, we built two levels of indexes, and search against both indexes for each topic. Then we combine these two searches into one. For Genomics track, we used a new strategy to automatically expand search terms by using relevance feedback and combined retrieval results from different fields into the final result. We achieved good results on the HARD task and average results on the Genomics task. For the HARD passage level evaluation, the automatic run `yorku04ha1' obtained the best result (0.358) in terms of Bpref measure at 12K characters. The evaluation results show that Algorithm 1 is more effective than Algorithm 2 for the passage level retrieval,

• Type: Conference
• Year: 2004
• Where: TREC
• Authors: Xiangji Huang, Yan Rui Huang, Ming Zhong, Miao Wen
• Keywords: Genomics Track, HARD Passage Level, Passage Level, TREC 2004, TREC 2008