Towards the Discovery of Semantic Relations in Large Biomedical Annotated Corpora

This paper proposes the application of multidimensional analysis over large semantically annotated biomedical corpora for the identification of relevant abstract relations between the recognized entities. The identification of relations is one of the most challenging issues in information extraction, as they guide the definition of the patterns used during the extraction phase. Multidimensional analysis allows us to define different analysis perspectives with different detail levels over the extracted facts. Among other tasks, users can distinguish discriminative relation patterns from ambiguous ones, detect the most relevant relation patterns and identify clusters of patterns that can refer to the same abstract relation. The proposal has been implemented upon a commercial tool and tested over the CALBC corpus.