MeSH Based Feedback, Concept Recognition and Stacked Classification for Curation Tasks

This paper reports about experiments carried out in the context of the genomics track at TREC 2004. Experiments were concentrated on two subtasks: the ad hoc retrieval task and the triage task. Experiments for the ad hoc task aimed at improving a standard full-text ad-hoc run (using a language modeling approach) by exploiting the manual classification of MEDLINE abstracts (the MeSH terms) for relevance feedback. The triage task was modeled as a standard classification task, using stacked classifiers and complex features, recognized by the Collexis IR engine.