Exploring structural features in predicting social network evolution

In this paper, we present a novel approach to incorporate the activity features in measuring the influence of member activities on the social network evolution. Conventional methods analyze social networks and make predictions based on all cumulative members and activities. However, since inactive members do not contribute to the network growth, including them in analysis can lead to less accurate results. Based on this observation, we propose to focus on the active population and explore the influence of member activities. We present a model that can incorporate various activity features and predict the evolution of the social activity. At the same time, an algorithm is adopted to select the most influential activity features. The experiments on two different types of social network show that the activity features can predict the evolution of the social activity accurately and our algorithm is effective to select the most influential features. Additionally, we find that the most significant activity features to determine the network evolution vary among different types of social network.