Extraction of Document Structure for Genomics Documents

We are taking as our foundational assumption that effective information retrieval tasks in the broad domain of biomedical literature must address the singular nature of scholarly communication and the effect this has upon a document corpus. The corpus for a typical TREC task is comprised of a temporal sequence of news documents exhibiting little, if any, internal structure. Reduction of a document to a single term vector is viable because of that document’s inherent aboutness regarding the story being reported. A scholarly paper, on the other hand, has distinct components, each fulfilling a specific function and exhibiting a correspondingly specific syntactic structure. A query to retrieve documents reporting new results regarding a particular biological organism should not retrieve candidates that mention that organism in the references or background sections, but rather ones that mention the organism in the results and discussion sections.