Follow you from your photos

In this work, we focus on the travelling location prediction problem of detecting whether a person will leave his living area and where he will go by analyzing the hidden connection between the user behaviors on geography and online social interactions. By analyzing more than 40, 000 Instagram media records from 26, 000 users, spanning a period of 3 months, we give special consideration to rarely visits locations, which are often ignored as noise in previous works, and we employ the dynamic Bayesian network to estimate the users' behavior and predict the location according to a majority voting model based on the social interaction information. We compare our model on the data of Instagram with two existing location prediction models, and find that (1) our model performs well both in the general location prediction and the location outside the living area.(2) social ties are effective for solving the location prediction problem as the accuracy of the prediction gets higher, given more social interaction information.