Coco model: A predictive model to explain interaction phenomena in social networks

Online social networks have provided an effective and revolutionary approach to facilitate the information diffusion in recent years. And it is a fundamental process taking place incessantly that individuals are making decisions to retweet or ignore microblog contents exposed from their following friends. Based on the real dataset, our research shows that the adoption probability would decline over time and is influenced by interactions with other topic contents. In this paper, we propose a retweet probabilistic model based on the network structure in both spatial and temporal dimensions, and derive a predictive model, named Coco model. We simplify Coco model in a specific situation where a couple of microblogs show strong interaction mutually and other external factors could be ignored. We evaluate and compare it with several previous models in term of fitting and predicting, on the basis of the real dataset collected from Weibo which includes 1,263,988 microblog contents. Our experiment results show that the Coco model could characterize the diffusion process and explain the interaction phenomenon well, and has outstanding prediction precision. For example, the average prediction accuracy of the Coco model for 30 days is 93.58% based on hotspots from Aug. 2014 to Dec. 2015.