Data science methods in political science research: analyzing protest activity in social media

The advent of social media and increased digitization of social processes have had a dramatic impact on politics and, particularly, on political mobilization and communication. The political science methodology and toolkit have also adapted to these changes and absorbed a variety of new approaches and methods from the burgeoning field of data science. This paper provides an overview of some of the key methodological innovations to the political science toolkit drawn from data science and discusses the advantages and limitations of these new methods for studying protest activity and political mobilization in social media. We focus on supervised and unsupervised learning as two major groups of methods that can be applied to either facilitate data collection in almost real time or the analysis of big data on protest activity. We discuss overfitting, regularization, and hyperparameter selection via cross-validation in the context of supervised methods, and present topic modeling and social network analysis techniques within unsupervised methods. The strengths and weaknesses of these methods are illustrated with references to recent articles published in peerreviewed journals. We conclude the paper with a discussion of the emerging methods that have not been used in political mobilization research yet and are open for further exploration by political scientists.