Emotion Classification in Russian: Feature Engineering and Analysis

Abstract

In this paper, we address the issue of identifying emotions in Russian informal text messages. For this purpose, a new large dataset of text messages from the most popular Russian messaging/social networking services (Telegram, VK) was compiled semi-automatically. Emojis contained in the text messages were used to annotate the data for emotions expressed. This paper proposes an integrated approach to text-based emotion classification combining linguistic methods and machine learning. This approach relies on morphological, lexical, and stylistic features of the text. Furthermore, the level of expressiveness was considered as well. As a result, an emotion classification model demonstrating near-human performance was designed. In this paper, we also report on the importance of different linguistic features of the text messages for the task of automatic emotive analysis. Additionally, we perform error analysis and discover ways to improve the model in the future.

RuSentiTweet: a sentiment analysis dataset of general domain tweets in Russian

“…Given the significant Russian-speaking population and the ever-growing level of Internet penetration, texts published by Russian-speaking users on social networks are attracting more and more attention from researchers. As a result, every year new works appear both in the classical analysis of the sentiment of Russian-language content ( e.g ., Araslanov, Komotskiy & Agbozo, 2020 ; Kanev et al, 2022 ; Kausar, Soosaimanickam & Nasar, 2021 ) and in related areas, such as the identification of emotions ( e.g ., Babii, Kazyulina & Malafeev, 2020 ; Kazyulina, Babii & Malafeev, 2020 ; Babii, Kazyulina & Malafeev, 2021 ), toxicity and hate speech detection ( e.g ., Zueva, Kabirova & Kalaidin, 2020 ; Pronoza et al, 2021 ; Smetanin & Komarov, 2021b ), and inappropriate language identification ( e.g ., Babakov et al, 2021 ; Babakov, Logacheva & Panchenko, 2022 ).…”