The deployment of new technologies in digital devices and communications has increased social networks' popularity and pervasiveness. Geo-tagged content has connected cyberspace to the real world by adding a new dimension to social networks. Among geo-tagged content available on social networks, geo-tagged images show users' communication and interaction with the environment in a better way. Because users frequently photograph regions of interest, these images can be used in many applications, including discovering regions of interest. Compared to traditional methods such as censuses and surveys, geo-tagged images benefit from saving time and expense to discover and analyze regions of interest. Therefore, researchers can use them in urban management and tourist recommendation. The purpose of this study is to discover the region of interest using geo-tagged data. Also, extracting appropriate semantic information and analysis in different contexts to identify regions of interest and understand the reason for their attractiveness is another goal of this research. This paper uses the Flickr geo-tagged images taken from New York City between 2015 and 2018. In the preprocessing phase, noise and data redundancy was removed. Then the data were clustered by the HDBSCAN method, and adjacent clusters that were similar in terms of text tags were merged. As a result, 106 regions of interest were identified. At the next step, a concave surface was fitted to points by α-shape method, and semantic information including distinguished labels, names, and categories, was selected for regions of interest. Finally, attractive regions were analyzed based on the type of visitors, users’ sentiment and the number of visits in different contexts. The evaluation results show the discovery of regions of interest in different shapes, dimensions, and densities. Our result corresponded for 66% with TripAdvisor's top attractions, while for the simple DBSCAN method this value was 53%. In regions that overlapped with TripAdvisor attractions, the naming was 76% similar.