Exploring Structural Analysis of Place Networks Using Check-In Signals Xiang Ding Department of Computer Science University of Massachusetts Lowell 1 University Avenue, Lowell, MA 01854 xding@cs.uml.edu Jing Xu Department of Computer Science University of Massachusetts Lowell 1 University Avenue, Lowell, MA 01854 jxu@cs.uml.edu Guanling Chen Department of Computer Science University of Massachusetts Lowell 1 University Avenue, Lowell, MA 01854 glchen@cs.uml.edu Abstract—The huge amount of check-in data obtained through location-based social networks (LBSNs) provides a great oppor- tunity to learn the characteristics of a geographic area through the users’ collective check-in behavior. In this paper, we explore structure analysis of place networks, in which vertices are geographic places while the links between places are formed based on the user’s check-in history. Specifically, we apply Louvain community detection algorithm on the place networks to obtain a set of place clusters (or communities). We found that these communities can be used to discover geographic layouts in the city and can uncover interesting patterns by analyzing geographically distant places within the same community. These results suggest that structural analysis of place networks is a promising approach for urban computing and has many implications of mobile networking, urban planning, user profiling and travel recommendations. Index Terms—location based social networks, social network analysis, community detection I. I NTRODUCTION In recent years we have witnessed the thriving rise of location-based social networks (LBSNs) driven by the rapid proliferation of smartphones. The mobile nature of these devices promotes the use of these social networks anytime, anywhere, thereby generating a vast amount of human be- havioral data with rich semantics that brings a new set of opportunities for researchers and application developers. There is a rich literature of structural analysis of the user networks of these LBSNs to analyze user friendships and mobility patterns [1]–[7]. Unlike these studies, we propose to apply structural analysis for Place Networks, in which vertices are geographic places and edges are the links between these places. The formation of these links depends on the user’s mobility patterns. Namely, we view the places a user visits in everyday life are inherently linked and we believe applying structural analysis methods to place networks can uncover novel insights of the relationship among places. This new knowledge can have many applications, such as urban planning, business analytics, user profiling, and travel recommendations. In this paper, we create place networks using Foursquare’s check-in data and apply community-detection algorithms to cluster the places. To balance location granularity and data density, here a place is a small geographic local subarea and a link between two places represents consecutive check-ins in these two places (one check-in in each place). The obtained place clusters, or called “communities”, matched well with the geographic layout of cities (Manhattan area of New York City in our study). By further analyzing the community members, we found that the same-community members may not always be geo- graphically close. Some places can be far away from other places even though they are in the same community (cluster). This is a bit counter-intuitive as people tend to move in small regions. By inspecting these distant community members, we uncovered several interesting insights, such as special events, commuting patterns, and user interests. These results show that structural analysis of place networks is a promising approach to obtain new knowledge of the cities and user behaviors. By continuous tracking the place networks, it is a much efficient approach to identify the evolutionary changes of the city life. The rest of this paper is organized as follows. We discuss the related work in Section 2. In Section 3 we describe how we collected Foursquare check-in data, based on which we created place networks and calculated place clusters using a community-detection algorithm. The analysis results are presented in Section 4. We conclude and discuss future work in Section 5. II. RELATED WORK Many researchers have exploited the rich information gen- erated by location-based social networks (LBSNs), such as to infer friendships [1], [2], to predict friendship links [3], [4], and to uncover human mobility patterns [5]–[7]. Our work differs with these approaches by focusing on place networks instead of friend networks. Recently some researchers have started to analyze properties of geographic locations using LBSN data. For example, in [8] the authors modeled human activity and geographical areas to compare urban neighborhoods within and across cities. In [9], the authors used geolocated tweets as a complementary source to determine land uses and urban points of interests. In [10], the authors applied a spectral clustering algorithm to find dynamic city areas which they call Livehoods. Our work differs with these approaches by clustering over explicit graph