Incremental Learning for Interaction Dynamics with the Influence Model

social networks, which are webs of relationships growing from computer-mediated interactions, have been explored in a wide variety of application domains, such as collaborative infor- mation recommendation, collective decision-making, viral mar- keting plan, etc. In these cases, it is crucial to understand how the networks dynamically affect the users' behaviors. This paper refers to the evolving influence of social networks in the interaction processes as interaction dynamics, and proposes a probabilistic framework for it based on the Influence Model. Moreover, the paper presents a gradient-based algorithm to incre- mentally learn the model from time-series interaction data, and shows its abilities to characterize chain dependencies through simulation experiments on synthetic data. We also apply the model to mine the dynamic influence networks from the know- ledge-sharing sites. The experimental results demonstrate that the proposed model and its learning algorithm can effectively capture the inter-influence relationships between users, and thus can drive the network toward one that has a higher profit potential.