Abstract: Web-based citizen feedback systems have become commonplace in cities around the

world, resulting in vast amounts of data. Recent advances in machine learning and natural

language processing enable novel and practical ways of analysing it as big data. This paper

reports an explorative case study of sentiment analysis of citizen feedback (in Finnish) by

means of annotation with custom categories (Positive, Neutral, Negative, Angry, Constructive

and Unsafe) and predictive modelling. We analyse the results quantitatively and qualitatively,

illustrate the benefits of such an approach, and discuss the use of machine learning in the

context of studying citizen feedback. Custom annotation is a laborious process, but it offers

task-specific adaptation and enables empirically grounded analysis. In this study, annotation

was carried out at a moderate scale. The resulting model performed well in the most frequent

categories, while the infrequent ones remained a challenge. Nonetheless, this kind of approach

has promising features for developing automated systems of processing textual citizen feedback.