Inferring activities from social media data

Social media produce an unprecedented amount of information that can be extracted and used in transportation research, with one of the most promising areas being the inference of individuals' activities. Whereas most studies in the literature focus on the direct use of social media data, this study presents an efficient framework that follows a user-centric approach for the inference of users' activities from social media data. The framework was applied to data from Twitter, combined with inferred data from Foursquare that contains information about the type of location visited. The users' data were then classified with a density-based spatial classification algorithm that allows for the definition of commonly visited locations, and the individual-based data were augmented with the known activity definition from Foursquare. On the basis of the known activities and the Twitter text, a set of classification algorithms was applied for the inference of activities. The results are discussed according to the types of activities recognized and the classification performance. The classification results allow for a wide application of the framework in the exploration of the activity space of individuals.