A multilevel searching and re-ranking framework for information retrieval
Miao Wen, Xiangji Huang · 2006
In this paper, we propose a multilevel searching and re-ranking framework (MLSRF) for information retrieval. In this framework, indexes are constructed at multi-level such as document level and passage level. For each topic, we search against each index independently and generate a sub-result for corresponding index. The final result is generated by re-ranking and integrating the sub-results for all the indexes. Experiments on 2004 and 2005 HARD data sets show that the proposed framework can make a positive contribution on both document level retrieval and passage level retrieval. We also investigate the effect of the blind feedback method on TREC HARD data sets. The results show that the performance of blind feedback material depends on the topic it works on.