Social network user influence sense-making and dynamics prediction

• An evaluation framework to systematically measure the characteristics of the influence.
• A dynamic information propagation model to predict the influence dynamics of social network users.
• A comprehensive empirical study on a large-scale twitter dataset to compare the influence metrics.

Identifying influential users and predicting their “network impact” on social networks have attracted tremendous interest from both academia and industry. Various definitions of “influence” and many methods for calculating influence scores have been provided for different empirical purposes and they often lack the in-depth analysis of the “characteristics” of the output influence. In addition, most of the developed algorithms and tools are mainly dependent on the static network structure instead of the dynamic diffusion process over the network, and are thus essentially based on descriptive models instead of predictive models. Consequently, very few existing works consider the dynamic propagation of influence in continuous time due to infinite steps for simulation. In this paper, we provide an evaluation framework to systematically measure the “characteristics” of the influence from the following three dimensions: (i) Monomorphism vs. Polymorphism; (ii) High Latency vs. Low Latency; and (iii) Information Inventor vs. Information Spreader. We propose a dynamic information propagation model based on Continuous-Time Markov Process to predict the influence dynamics of social network users, where the nodes in the propagation sequences are the users, and the edges connect users who refer to the same topic contiguously on time. Finally we present a comprehensive empirical study on a large-scale twitter dataset to compare the influence metrics within our proposed evaluation framework. Experimental results validate our ideas and demonstrate the prediction performance of our proposed algorithms.