Event prediction with community leaders

Abstract—With the emerging of online social network services, quantitative studies on social influence become achievable. Lead-ership is one of the most intuitive and common forms for social influence; understanding it could result in appealing applications such as targeted advertising and viral marketing. In this work, we focus on investigating leaders ’ influence for event prediction in social networks. We propose an algorithm based on events that users conduct to discover leaders in social communities. Analysis on the leaders that we found on a real-life social network dataset leads us to several interesting observations, such as that leaders do not have significantly higher number of friends but are more active than other community members. We demonstrate the effectiveness of leaders ’ influence on users ’ behaviors by learning tasks: given a leader has conducted one event, whether and when a user will perform the event. Experimental results show that with only a few leaders in a community the event predictions are always very effective. I