Mining the Web for Sympathy: The Pussy Riot Case

With social media services becoming more and more popular, there now exists a constant stream of opinions publicly available on the Internet. In crisis situations, analysis of social media data can improve situation awareness and help authorities to provide better assistance to the affected population. The large amount of activity on social media services makes manual analysis infeasible. Thus, an automatic system that can assess the situation is desirable. In this paper we present the results of training machine learning classifiers to being able to label tweets with one of the sentiment labels positive, neutral, and negative. The classifiers were evaluated on a set of Russian tweets that were collected immediately after the much debated verdict in the 2012 trial against members of the Russian punk rock collective Pussy Riot. The aim for the classification process was to label the tweets in the dataset according to the author´s sentiment towards the defendants in the trial. The results show that the obtained classifiers do not accurately and reliably classify individual tweets with sufficient certainty. However, the classifiers do show promising results on an aggregate level, performing significantly better than a majority class baseline classifier would.