Skip to Main Content
In this paper, we propose examining the participants in various meetings or communications within a social network, and using sequential inference based on these participant lists to quickly and accurately predict anomalies in the content of those communications. The proposed approach consists of two main elements: (1) filtering, or assigning a belief or likelihood to each successive measurement based upon our ability to predict it from previous noisy observations, and (2) hedging, or flagging potential anomalies by comparing the current belief against a time-varying and data-adaptive threshold. The threshold is adjusted based on feedback requested from an expert system. In general, parsing communication data can require nontrivial computational resources, but since parsed data is only used sparingly for feedback, the overall computational complexity of the proposed approach is relatively low. Regret bounds quantify the performance of the proposed approach, and experiments on the Enron email database demonstrate its efficacy.