Identifying Social Network Patterns with Exponential Random Graph Models

Abstract Social networks remain central to successful multi-domain operations, and yet key dimensions of network structure remain understudied. Network motifs represent local structures, such as dyads, triads, and tetrads that are overrepresented. Here, we develop and apply novel methodological techniques for analyzing how these subgraphs define social networks. First, we examine the prevalence of motifs among 24 networks that represent six types of social interactions: friendship, legislative co-sponsorship, Twitter messages, advice seeking, email communication, and terrorist collusion. Using exponential random graph models (ERGMs), we find important commonalities across networks and document five motifs (one dyad, three triads, and one tetrad). Networks display unique “signatures” that can be used to identify the type of network involved in interchanges. Next, we apply this ERGM approach to compare patterns in positive (e.g., cooperative) and negative (e.g., conflictual) networks and uncover key differences. Finally, we use AI/ML techniques to predict the spread of negative, online network communications over time. Our findings demonstrate the utility of state-of-the-art statistical techniques to investigate network patterns, both at one point in time, and longitudinally. Applying an innovative, multivariate, ERGM approach to detecting motifs has distinct advantages over previous procedures. Using AI/ML is useful in predicting the dissemination of negative, online interchanges.
  • Diane Felmlee (Penn State)
  • Cassie McMillan (Penn State)
  • Roger Whitaker (Cardiff)
  • Mudhakar Srivasta (IBM US)
  • Cheryl Giammanco (ARL)
  • Emma Jayes (Dstl)
Date Sep-2021
Venue Technical report