RobBERT finetuned for sentiment analysis on DBRD

This is a finetuned model based on RobBERT (v2) . We used DBRD , which consists of book reviews from hebban.nl . Hence our example sentences about books. We did some limited experiments to test if this also works for other domains, but this was not exactly amazing.

We released a distilled model and a base -sized model. Both models perform quite well, so there is only a slight performance tradeoff:

*The results of RobBERT are of a different run than the one reported in the paper.

Training data and setup

We used the Dutch Book Reviews Dataset (DBRD) from van der Burgh et al. (2019). Originally, these reviews got a five-star rating, but this has been converted to positive (⭐️⭐️⭐️⭐️ and ⭐️⭐️⭐️⭐️⭐️), neutral (⭐️⭐️⭐️) and negative (⭐️ and ⭐️⭐️). We used 19.5k reviews for the training set, 528 reviews for the validation set and 2224 to calculate the final accuracy.

The validation set was used to evaluate a random hyperparameter search over the learning rate, weight decay and gradient accumulation steps. The full training details are available in training_args.bin as a binary PyTorch file.

Limitations and biases

• The domain of the reviews is limited to book reviews.
• Most authors of the book reviews were women, which could have caused a difference in performance for reviews written by men and women .
• This is not the same model as we discussed in our paper, due to some conversion issues between the original training two years ago and now, it was easier to retrain this model. The accuracy is slightly lower, but the model was trained on the beginning of the reviews instead of the end of the reviews.

Credits and citation

This project is created by Pieter Delobelle , Thomas Winters and Bettina Berendt . If you would like to cite our paper or models, you can use the following BibTeX:

@inproceedings{delobelle2020robbert,
    title = "{R}ob{BERT}: a {D}utch {R}o{BERT}a-based {L}anguage {M}odel",
    author = "Delobelle, Pieter  and
      Winters, Thomas  and
      Berendt, Bettina",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2020",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.findings-emnlp.292",
    doi = "10.18653/v1/2020.findings-emnlp.292",
    pages = "3255--3265"
}