Social media semantic perceptions on Madrid Metro system: Using Twitter data to link complaints to space

Social networks are platforms widely used by travelers who express their opinions about many services like public transport. This paper investigates the value of texts from social networks as a data source for detecting the spatial distribution of problems within a public transit network by geolocating citizens' feelings, and analyzes the effects some factors such as population or income have over that spatial spread, with the goal of developing a more intelligent and sustainable public transit service. For that purpose, Twitter data from the Madrid Metro account is collected over a two-month period. Topics and sentiments are identified from text mining and machine learning algorithms, and mapped to explore spatial and temporal patterns. Lastly, a Geographically Weighted Regression model is used to explore the causality of the spatial distribution of complaining users, by using official data sources as exploratory variables. Results show Twitter users tend to be mid-income workers who reside in peripheral areas and mainly tweet when traveling to workplaces. The main detected problems were punctuality and breakdowns in transfer stations or in central areas, mainly in the early morning of weekdays, and affected by density of points of interest in destination areas.