From Individual Behavior to Influence Networks: A Case Study on Twitter

Understanding social influence and its related phenomena is a major challenge in the study of the human collective behavior. In the recent years, the availability of internet-based communication and interactivity data has enabled studies on social influence at an unprecedented scale and time resolution. In this work, we study how individual behavior data may provide knowledge regarding influence relationships in a social network. We define what we call the influence network discovery problem, which consists of identifying influence relationships based on user behavior across time. Our objective is the design of accurate models that are able to exploit different types of behavior in order to discover how people influence each other. Several strategies for influence network discovery are proposed and discussed. Moreover, we present a case study on the application of such strategies using a follower-followee network and user activity data from Twitter, which is a popular microblogging and social networking service. We consider that a follower-followee interaction defines a potential influence relationship between users and the act of posting a tweet, a URL or a hashtag represents an individual behavior on Twitter. The results show that, while tweets may be used effectively in the discovery of influence relationships, hashtags and URLs do not lead to good performance in such task. Moreover, strategies that consider the time when an individual behavior is observed outperform those that do not and by combining such information with the popularity of the behaviors, even better results may be achieved.