Automatic detection of user trajectories from social media posts

Social media represents a rich environment to collect huge amounts of data containing useful information about people's behaviors and interactions. In particular, such information has been widely exploited for analyzing the mobility of people, as geotagged social media posts allow to extract accurate patterns on movements of people. This paper presents AUDESOME (AUtomatic Detection of user trajEctories from SOcial MEdia), an automatic method for discovering user mobility patterns from social media posts. In particular, the method includes two new unsupervised algorithms: (i) a text mining algorithm, which analyzes social media posts to automatically extract the main keywords identifying the Places-of-Interest (PoI) in a given area; and (ii) a geospatial clustering algorithm, which detects the Regions-of-Interest (RoIs) by using both geotagged posts and extracted keywords. We experimentally evaluated the performance of AUDESOME taking into account the following aspects: identification of keywords, detection of RoIs, and extraction of user trajectories. The experiments, performed on a real dataset containing about 3 million of geotagged items published in Flickr, demonstrate that AUDESOME achieves better results than existing techniques.