Sentiment Analysis for Tigrinya with TiRoBERTa
This model is a fine-tuned version of TiRoBERTa on a YouTube comments Sentiment Analysis dataset for Tigrinya (Tela et al. 2020).
Basic usage
from transformers import pipeline
ti_sent = pipeline("sentiment-analysis", model="fgaim/tiroberta-sentiment")
ti_sent("ድምጻዊ ኣብርሃም ኣፈወርቂ ንዘልኣለም ህያው ኮይኑ ኣብ ልብና ይነብር")
Training
Hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
Results
It achieves the following results on the evaluation set:
- F1: 0.8477
- Precision: 0.7607
- Recall: 0.957
- Accuracy: 0.828
- Loss: 0.6796
Framework versions
- Transformers 4.10.3
- Pytorch 1.9.0+cu111
- Datasets 1.10.2
- Tokenizers 0.10.1
Citation
If you use this model in your product or research, please cite as follows:
@article{Fitsum2021TiPLMs,
author={Fitsum Gaim and Wonsuk Yang and Jong C. Park},
title={Monolingual Pre-trained Language Models for Tigrinya},
year=2021,
publisher={WiNLP 2021/EMNLP 2021}
}
References
Tela, A., Woubie, A. and Hautamäki, V. 2020.
Transferring Monolingual Model to Low-Resource Language: The Case of Tigrinya.
ArXiv, abs/2006.07698.
- Downloads last month
- 13
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Evaluation results
- Accuracyself-reported0.828
- F1self-reported0.848
- Precisionself-reported0.761
- Recallself-reported0.957