Prevalent Co-visiting Patterns Mining from Location-based Social Networks

Spatial co-location mining is a key problem in urban planning and marketing. Current spatial co-location mining methods ignore the people who are related to the co-location patterns' instances, which results that the mining results are hard to explain and understand by the users. In this paper, we combine the theories of co-location mining and social networks analysis to mine a kind of special co-location patterns: Co-visiting patterns, which consider spatial information and social information at the same time. A co-visiting pattern is also a spatial feature set, whose instances are always visited by the similar users and located in a nearby region. We propose some new measures, including the user similarity, the weight of neighborhood relationship of two visited spatial instances, and the prevalent degree of a co-visiting pattern. In addition, we also explore the properties of the co-visiting patterns in this paper, and present an efficient algorithm. Finally, experiments and a detailed analysis are given at the end of this paper. Experimental results show that the rationality of co-visiting pattern, and the effectiveness and stability of the mining algorithm.