Hypothesis generation guided by co-word clustering

Co-word analysis was applied to keywords assigned to MEDLINE documents contained in sets of complementary but disjoint literatures. In strategical diagrams of disjoint literatures, based on internal density and external centrality of keyword-containing clusters, intermediate terms (linking the disjoint partners) were found in regions of below-median centrality and density. Terms representing the disjoint literature themes were found in close vicinity in strategical diagrams of intermediate literatures. Based on centrality-density ratios, characteristic values were found which allow a rapid identification of clusters containing possible intermediate and disjoint partner terms. Applied to the already investigated disjoint pairs Raynaud's Disease - Fish Oil, Migraine Magnesium, the method readily detected known and unknown (but relevant) intermediate and disjoint partner terms. Application of the method to the literature on Prions led to Manganese as possible disjoint partner term. It is concluded that co-word clustering is a powerful method for literature-based hypothesis generation and knowledge discovery.