Inferring Friendship From Check-In Data Of Location-Based Social Networks

In recent years, the relation between human mobility and social connection attracts more attention from both industry and academia. In the past, obtaining people’s mobility information is considered to be an obstacle for related study.Nowadays, however, with the ubiquity of GPS-enabled mobile devices and location-based social networking services, gaining people’s spatial-temporal location information becomes easier. These spatial-temporal location information brings us opportunities to investigate on some valuable information of a particular individual, such as his interests, his visiting pattern and his real-life social connections.Particularly, in this paper, we focus on predicting whether there exists friendship between two individuals according to their mobility information. As friends tend to visit same places, we consider the number of co-occurrences and the number of locations that two people co-occurred to be indicators of friendship. Besides, the visiting time interval between two users also has an effect on friendship prediction.By conducting machine learning technique on dataset abstracted from all the information above, we construct two friendship prediction models. The first model refers to predicting friendship of two people with only one of their co-occurred places’ information. We take different numbers of check-ins, characteristic of the location and check-in time interval between two users into consideration. Besides, we define the time interval sequence to describe the general check-in time interval of two users at one location. This model aims at achieving an effective prediction model under the situation that only limited resources are provided. The second model proposes solution for predicting friendship of two people where all the information of their every particular co-occurred place is essential. We utilize the number of co-locations, the number of co-occurrences, characteristic of locations and check-in time intervals. Two notions, weighted number of co-locations and weighted number of co-occurrences, are defined to combine location entropy and co-occurrences together to achieve a better result. The result shows that both our models outperform the state-of-art models for location-based friendship prediction.