Predicting Argumentative Influence Probabilities in Large-Scale Online Civic Engagement

Gaku Morio, Katsuhide Fujita

Abstract: Large-scale online civic engagements (OCEs) with more than 100 participants have become possible due to recent developments in online social media technology. OCEs have the potential to achieve consensus building and collective decision-making with a large number of citizens, which is difficult to achieve in face-to-face contexts. However, most users in a large-scale OCE are rarely constantly active. Therefore, an important problem for the activation of a large-scale OCE is to facilitate the discussion by predicting which citizens will have significant influence in the discussion. This paper examines the activation prediction problem in a large-scale OCE. We propose a novel influence model based on the impulse response of activity histories and argumentative pressures, as well as an effective testing algorithm. The experimental results demonstrate that the proposed models with impulse response and the lexical pressures show better accuracy compared with baselines. In addition, the testing time required by the proposed method can be reduced significantly by employing a node-cutting algorithm.