Inferring pairwise influence from encrypted communication

Inferring influence in networks from observed communication activity is an important task in contexts such as surveillance, marketing, and cybersecurity. Most existing approaches rely on content or meta-data indicating related activity, rendering those approaches ineffective when such information is unavailable, for example due to encrypted communication. In contrast, we present an efficient algorithm to infer influence between entities that relies only on the times of their individual activity, paying particular attention to the computational challenges posed by large, high-volume networks. We provide theoretical bounds on the recall and runtime of our algorithm relative to characteristics of network structure and dynamics, along with experiments to support our theoretical analysis and validate the effectiveness of our approach.