FastText-based methods for Emotion Identification in Russian Internet Discourse

Abstract

In this paper we tackle the problem of emotion detection and classification in Russian short text messages. We use such recent NLP development as fastText that produces state-of-the-art results for a variety of tasks. We put a special emphasis on the challenges that arise while using a dataset of text messages from the most popular Russian messaging/social networking services (Telegram, VK). We also provide an extensive quantitative prediction analysis along with suggestions of possible ways to improve the results. Finally, we discuss the prospects of developing and implementing discourse-specific emotion identification technologies for the Web.