This repository provides a source code and data for the paper "Transfer Learning for Improving Results on Russian Sentiment Datasets" (Anton Golubev and Natalia Loukachevitch, Dialogue 2021).

This repo provides only scripts for creating dataset of thematic contexts from raw corpus of news. The source code for the BERT models described in the article, you can find in https://github.com/antongolubev5/Targeted-SA-for-Russian-Datasets.

Requirements

• python: 3.7.1
• numpy: 1.15.4
• nltk
• sklearn
• pandas
• os
• tqdm
• xml
• time
• spacy
• gensim
• pymorphy2
• seaborn

References

@misc{golubev2021improvingtf,
    title={Transfer Learning for Improving Results on Russian Sentiment Datasets},
    author={Anton Golubev and Natalia Loukachevitch},
    year={2021},
    eprint={2007.14310},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}