Improving Results on Russian Sentiment Datasets
This work improves sentiment analysis for Russian language applications, but it is incremental as it applies existing methods to specific datasets.
The study tested standard neural networks and Russian BERT variants on Russian sentiment datasets, finding that a conversational BERT variant performed best, with the BERT-NLI model achieving near-human performance on one dataset.
In this study, we test standard neural network architectures (CNN, LSTM, BiLSTM) and recently appeared BERT architectures on previous Russian sentiment evaluation datasets. We compare two variants of Russian BERT and show that for all sentiment tasks in this study the conversational variant of Russian BERT performs better. The best results were achieved by BERT-NLI model, which treats sentiment classification tasks as a natural language inference task. On one of the datasets, this model practically achieves the human level.