Fig. 4From: Leveraging graph topology and semantic context for pharmacovigilance through twitter-streamsDEGSider feature distributions. The DEGSider topological feature distributions for Common Neighbor Size, Jaccard Coefficient, and Preferential Attachment (parts 1,2 and 3). In each case, the shape is distinct between the adverse and pseudo-non-adverse edges. Part 4 shows all features projected to the three most explanatory PCA components. As with individual features, each class takes a distinct shapeBack to article page