Large Sentiment Dictionary of Russian Words

Sentiment analysis is a widely studied area of computational linguistics. The main tool for sentiment analysis of texts are dictionaries with positive/negative ratings of words. Hundreds of such dictionaries have been compiled for dozens of world languages. The largest English dictionary contains up to 500 thousand words. By contrast, the largest dictionary of the Russian language contains less than 50 thousand words. We compiled a large Russian dictionary with positive/negative ratings of approximately 2 million word forms. To do this, we applied the technique of machine extrapolation of valence ratings contained in 6 existing dictionaries obtained by the survey method. Pretrained fasttext vectors were used as input to the neural network sentiment predictor. Multiple trained models allowed us for cross–validation of sentiment estimations. The obtained Spearmen‘s correlation coefficient between the human ratings and their machine estimates is up to 0.835.