Dynamic prediction of communication flow using social context

In this paper, we develop a temporally evolving representation framework for context that can efficiently predict communication flow in social networks between a given pair of individuals. The problem is important because it facilitates determining social and market trends as well as efficient information paths among people. We describe communication flow by two parameters: the intent to communicate and communication delay. To estimate these parameters, we design features to characterize communication and social context. Communication context refers to the attributes of current communication. Social context refers to the patterns of participation in communication (information roles) and the degree of overlap of friends between two people (strength of ties). A subset of optimal features of the communication and social context is chosen at a given time instant using five different feature selection strategies. The features are thereafter used in a Support Vector Regression framework to predict the intent to communicate and the delay between a pair of individuals. We have excellent results on a real world dataset from the most popular social networking site, www.myspace.com. We observe interestingly that while context can reasonably predict intent, delay seems to be more dependent on the personal contextual changes and other latent factors characterizing communication, e.g. 'age' of information transmitted and presence of cliques among people.