Vegetarianism Discourse in Russian Social Media: A Case Study

Dietary choices, especially vegetarianism, have attracted much attention lately due to their potential effects on the environment, human health, and morality. Despite this, public discourse on vegetarianism in Russian-language contexts remains underexplored. This paper introduces VegRuCorpus, a novel, manually annotated dataset of Russian-language social media texts expressing opinions on vegetarianism. Through extensive experimentation, we demonstrate that contrastive learning significantly outperforms traditional machine learning and fine-tuned transformer models, achieving the best classification performance for distinguishing pro- and anti-vegetarian opinions. While traditional models perform competitively using syntactic and semantic representations and fine-tuned transformers show promise, our findings highlight the need for task-specific data to unlock their full potential. By providing a new dataset and insights into model performance, this work advances opinion mining and contributes to understanding nutritional health discourse in Russia.