Enhancing the Impression on Cities: Mining Relations of Attractions with Geo-Tagged Photos

With mobile devices becoming increasingly popular in our daily life, geo-tagged data are widely available and shared through various social media platforms. By exploring this type of data, much progress has been made by scholars to provide better location-based services for users. This paper aims to help people enhance their impression of cities with a dataset collected from Flickr, a photo sharing social media platform. First, to mine the pattern that users frequently move within the span distance of a city, we extract cities from raw data with Wave Cluster algorithm. Second, to find the points of interest (POIs) in each city, we cluster user sharing locations considering the accuracy of GPS technology. Next, to discover the tourist attractions and their relations, we build visit sets for each user. Visualization of relations on a digital map is also provided to users for better exploring the space of cities. In order to prove the practicality of relations we detect, attraction recommendation is performed and compared with HITS, an algorithm for attraction recommendation in the previous study. Our evaluation experiments on cities and attractions show that our results are highly consistent with the reality and superior to the HITS algorithm used in the previous study for attraction recommendation.