A Text Mining Approach to Enable Detection of Candidate Risk Factors

To manage information overload users segment their information resources into smaller and smaller partitions. Yet, these partitions are often orthogonal to the expression of an underlying phenomenon. Our objective is to enable a user to identify new phenomena by developing automated methods that synthesize typically unused evidence from existing scientific articles. To demonstrate our approach, we have developed a system that combines existing meta-analytic techniques with a new way to partition an information space. Our preliminary experiments revealed statistically significant associations from a corpus of articles that were consistent with the entire literature. Furthermore, existing meta-analytic techniques were unable to detect from the same corpus.