Towards understanding traveler behavior in Location-based Social Networks

Abstract

Understanding users' behavior in Location-based Social Networks (LBSNs) is becoming an interesting research topic. In LBSNs, users can explore the places of interest around their current locations, check in at these locations and share such check-ins with their friends or the public. Therefore, the check-ins are valuable information for studying user behavior. Many services would benefit from the research of user behavior. For example, it can help the urban design and development based on the user mobility patterns and it could also improve the location recommendations to help users to find their places of interests. Intrinsically, traveler's activities in LBSN are distinctive, especially when compared with the a local user's activities. Therefore, a study of travelers' activities in LBSNs can help understand traveler behavior and then help LBSNs provider to improve their services, e.g. location recommendation service to new visitors. The location recommendation is especially important for a new visitor to a city. However, in the literature, there is little work specially focusing on the research of travelers' behavior in LBSNs. In this paper, we take the first step towards understanding such user behavior in LBSNs. Our research is based on the travelers' check-in information created in the greater Pittsburgh area in Foursquare. At first, we empirically study the venues and the check-ins created on such venues based on venue category information. After that, we investigate the temporal features of travelers' check-ins, and examine the evolution of check-ins created at the venues related to four categories using spatio-temporal information. Besides the empirical study, we employ the notion of user entropy to investigate the diversity of the travelers' check-ins. Through the research of the user entropy as a function of the user's check-ins, we find that the majority travelers usually exhibit higher diversity in their activities. Moreover, we also use the Latent Dirichlet Allocation (LDA) to generate travelers' mobility patterns. These human centric latent topics cannot only help to cluster the venues but also address the hot spots in a city based on the crowd level.