kinit/slovakbert-sentiment-twitter

Sentiment Analysis model based on SlovakBERT

This is a sentiment analysis classifier based on SlovakBERT. The model can distinguish three level of sentiment:

• -1 - Negative sentiment
• 0 - Neutral sentiment
• 1 - Positive setiment

The model was fine-tuned using Slovak part of Multilingual Twitter Sentiment Analysis Dataset [Mozetič et al 2016] containing 50k manually annotated Slovak tweets. As such, it is fine-tuned for tweets and it is not advised to use the model for general-purpose sentiment analysis.

Results

The model was evaluated in our paper [Pikuliak et al 2021, Section 4.4]. It achieves $0.670.67$ F1-score on the original dataset and $0.580.58$ F1-score on general reviews dataset.

Cite

@inproceedings{pikuliak-etal-2022-slovakbert,
    title = "{S}lovak{BERT}: {S}lovak Masked Language Model",
    author = "Pikuliak, Mat{\'u}{\v{s}}  and
      Grivalsk{\'y}, {\v{S}}tefan  and
      Kon{\^o}pka, Martin  and
      Bl{\v{s}}t{\'a}k, Miroslav  and
      Tamajka, Martin  and
      Bachrat{\'y}, Viktor  and
      Simko, Marian  and
      Bal{\'a}{\v{z}}ik, Pavol  and
      Trnka, Michal  and
      Uhl{\'a}rik, Filip",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2022",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.findings-emnlp.530",
    pages = "7156--7168",
    abstract = "We introduce a new Slovak masked language model called \textit{SlovakBERT}. This is to our best knowledge the first paper discussing Slovak transformers-based language models. We evaluate our model on several NLP tasks and achieve state-of-the-art results. This evaluation is likewise the first attempt to establish a benchmark for Slovak language models. We publish the masked language model, as well as the fine-tuned models for part-of-speech tagging, sentiment analysis and semantic textual similarity.",
}