HMM based cross-OSN user migration modeling and forecasting

With the incredible growth of OSNs (online social networks), users have numerous choices every moment. However, due to the limit of time and resources, only a small part of OSNs are chosen to remain social and active by users. The dynamic changes of users' interests entail user migration. Understanding user migration behavior is important to improve business intelligence and retain users. In this paper, a Hidden Markov Model (HMM)-based method is proposed to model user migration process. First, we take personal, neighbors' and websites' factors into account to define user migration states, and then use them as observation states to generate the observation sequences in the time span of the dataset. Second, to describe user migration process more accurately, these sequences are grouped into clusters by K-means, and we use Baum-Welch algorithm to train HMM parameters for every cluster. Third, users' future behaviors are predicted by the trained model using Maximum a Posteriori (MAP) approach. Finally, we carry out our study on real data sets; the result shows the accuracy can be improved to as high as 92.4% under the most ideal circumstance.