Computational Investigation of Linguistic Markers in Discourse of Political Adversaries via Interpretation of Recurrent Neural Network

A community discourse can be analyzed through texts written by participants of the community since it is expressed in those texts in various ways. For large communities, the sheer amount of texts generated limits the ability of human researcher to comprehend unique features of a discourse. But modern Machine Learning algorithms are able to process large amounts of text thus aiding the human researcher in investigation. In this study we offer: 1. A large corpus made from three types of text: writings of Russian pro-government and opposition activists and neutral texts without political coloring; 2. A modern word-level Recurrent Neural Network-based approach for unsupervised detection of discourse-specific linguistic