An online updating multi-threshold method for mobility relationship mining

With the popularity of the mobile phones and location-based social networks, rich location data has become widely available nowadays, enabling study on friendship detection based on human mobility. However, in some circumstances, limited by data collection techniques, only discrete location (such as location IDs) can be fetched which leads to methods of detecting friendship based on distance metric is unrealistic. This paper aims to detect friendship among users based on their mobility data where locations are represented by location IDs. By considering each user's preference for location and each location's popularity, the difference between friends and strangers gets larger than using merely frequency of meeting events. However, we discover that measure values obtained may vary greatly among different users and propose to maintain a threshold for each user. In addition, an online updating algorithm is presented to update thresholds when a new user's mobility data is available. Experiments conducted on MIT Reality Mining project dataset show that the proposed updating method is practical and the proposed multi-threshold model outperforms the state-of-the-art methods proposed by Wang et al.