Résumé
Social media data is an efficient means to understand human mobility through the spatial and temporal patterns of the users. Those patterns can help us discover and define mobility communities, which we specify as a group of users sharing the same spatiotemporal patterns. In this paper, we focus on a particular social media platform, X, formerly known as Twitter. X features geolocalized posts, also known as geolocalized tweets, that can be gathered through the platform's API. Our goal is to gather and analyze geolocalized tweets from two different cities, namely Brasilia and London, over a one-year period, to extract common spatiotemporal patterns among users and carry out a comparative analysis between cities. Social media data can be labeled as sparse data relative to other datasets. However, such data also offers the benefits of: (1) not being influenced as an object of a foreseen or tailored study; and (2) providing naturally both a textual contextualization and GPS geolocalization. Our goal is to exploit such assets toward human mobility pattern detections. In this paper, we show how two different methods, namely K-means clustering based on a user's vectorized text and the Louvain community detection algorithm on a user graph based on similarity score, can be used successfully to extract mobility patterns. We provide two case studies using London and Brasilia data for the year 2022 and carry out a comparative analysis of our methodology, highlighting common pattern behaviors.