Characterization of Social Land use in Urban Environments Based on the Semantic Dimension of Location Based Social Network's Data

Recognizing the urban environments and understanding the citizens’ motion behavior is an important research field in the area of spatial data analysis. The location-based social networks record and gather update, rich, and enormous data that users share them honestly through their spatial, temporal and semantic behavior. Undoubtedly the physical structure of an urban area as well as its land use impress the spatial behavior of its citizens and this impression propagates to the data of location-based social networks. Because of that, nowadays, researchers use the users’ data in location-based social networks in order to recognize urban environments. In this research, we attempted to cluster urban environments based on social land uses by using the location-based social networks’ spatial and semantic data. In this regard, in the first step, the spatial data of users are clustered by employing a clustering algorithm that is based on a competitive neural network (SOM). To cluster the spatial data of users, we first should calculate the optimal number of clusters. In this regard, Elbow chart was used as DB index. Then, the urban environment is partitioned into several regions by drawing the Voronoi diagram on the cluster centers and the data which users have been recorded in each region are identified. The number of data available in each region was computed for semantic categories separately, then the vector of each region was normalized. Similarly, these operations were repeated for all data in whole urban environment and the. The initial idea is usage of the abundance of each category of semantic data; however, this criterion cannot determine the land use of a region properly; because it is possible that users share more information about, for example, creation places than residential ones. Finally after extracting the percentage of the different groups of semantic data and by considering the weight of each group, a semantic dimension that is the representative of the region’s social land use was assigned to each region by taking advantage of a clustering algorithm based on the semantic dimension of users’ data. To evaluate the proposed method, the number of data in each category was calculated for every 15 minutes of a day to verify the validity of data that users share about their activities in the foursquare social network. To more accurate study, the working days and weekend days were studied separately; i.e. for each category, we formed a vector with 192 members. The chart of temporal variations of data numbers during a day (24 hours) was plotted for clusters identified from proposed method too. Then, the correlation among these charts was used as the evaluation index of the proposed method. This research and the performed evaluation show that the big data of social networks are not only low cost and updated but also shared by citizens honestly and have suitable validity. Also, the urban regions with common or similar social land uses have spatial continuity. The results of the research show the high potential of the location-based social networks to recognize urban environments.