Identifying groups of interest through temporal analysis and event response monitoring

We present a method for finding groups of interest in large social networks. A temporal group detection algorithm identifies tightly connected groups by analyzing communications as they unfold over time. Since the number of groups found through temporal group detection may be too large to allow for manual analysis of their behavior, we also present an algorithm to identify groups of interest within an existing set of groups. This algorithm works by observing how groups react to key events and finds groups of interest by noting which groups respond anomalously. We demonstrate this approach on two social media datasets collected from Twitter. The first dataset involves tweets from Afghanistan when Afghanistan signed a letter of cooperation with India. The second dataset involves tweets surrounding the death of Steve Jobs. In both cases, our algorithm was able to identify appropriate groups.