SNARE

摘要: Classifying nodes in networks is a task with wide range of applications. It can be particularly useful anomaly and fraud detection. Many resources are invested the detection due to high cost fraud, being able automatically detect potential quickly precisely allows human investigators work more efficiently. data analytic schemes have been put into use; however, that bolster link analysis prove promising. This builds upon belief propagation algorithm for use detecting collusion other schemes. We propose an called SNARE (Social Network Analysis Risk Evaluation). By allowing one domain knowledge as well knowledge, method was very successful pinpointing misstated accounts our sample general ledger data, significant improvement over default heuristic true positive rates, lift factor up 6.5 (more than twice heuristic). also apply graph labeling on publicly-available datasets. show only some information about themselves network, we get surprisingly accuracy labels. Not applicable variety domains, but it robust choice parameters highly scalable-linearly number edges graph.