Fast Prediction for Criminal Suspects through Neighbor Mutual Information-Based Latent Network

Figure 3

The latent network and the latent vectors. (a) A criminal network with the latent network composed of centroids of the case clusters. Colors represent cluster membership and squares represent cluster centroids. (b) The relative location estimation in circumstances of clusters of different shapes and densities. is estimated to be near the core of cluster 1 because all three of its neighbors belong to cluster 1 (homogeneous in terms of cluster membership), whereas or is estimated to be near the boundaries of clusters because the neighbors differ in cluster membership (heterogeneous). (c) The cluster memberships of and , NMI calculation result, and their latent vector .

(a)

(b)

(c)