Fusion of knowledge-intensive and statistical approaches for retrieving and annotating textual genomics documents

Abstract

This paper represents a continuation of research into the retrieval and annotation of textual genomics documents (both MEDLINE ® citations and full text articles) for the purpose of satisfying biologists’ real information needs. The overall approach taken here for both the ad hoc retrieval and categorization tasks within the TREC genomics track in 2005 was one combining the results of several NLP, statistical and ML methods, using a fusion method for ad hoc retrieval and ensemble methods for categorization. The results show that fusion approaches can improve the final outcome for the ad hoc and the categorization tasks, but that care must be taken in order to take advantage of the strengths of the constituent methods.