Datasets:
language: pt
license: apache-2.0
Dataset
A manually annotated dataset was created to enable supervised training for the FinBERT-PT-BR model, which focuses on sentiment analysis of Brazilian Portuguese financial texts. More than 1.4 million financial news texts in Portuguese were collected and used for the initial language modeling phase. From this corpus, a sample of 1,000 texts was manually annotated with sentiment labels.
Annotation Process
Three annotators participated in the process.
All texts were annotated by at least two different annotators.
The defined categories were: POSITIVE, NEGATIVE, NEUTRAL, and NOT_APPLICABLE.
Annotation guideline:
"Classify the text based on whether it would imply a Positive, Negative, or Neutral return. Use 'Not applicable' for texts unrelated to finance, involving politics, or nonsensical content."
After a calibration step to ensure agreement among annotators, the full sample was annotated.
Final Dataset Composition
Out of the 1,000 annotated texts:
497 texts were discarded due to lack of agreement or being labeled as “Not applicable”.
The final training dataset contains 503 texts:
160 positive
203 negative
140 neutral
Agreement Metrics
Agreement rate: 90.4%
Krippendorff’s alpha: 0.88
These metrics indicate a high level of consistency among annotators, validating the quality of the dataset used for supervised fine-tuning.
Author
Citation
@inproceedings{santos2023finbert,
title={FinBERT-PT-BR: An{\'a}lise de Sentimentos de Textos em Portugu{\^e}s do Mercado Financeiro},
author={Santos, Lucas L and Bianchi, Reinaldo AC and Costa, Anna HR},
booktitle={Anais do II Brazilian Workshop on Artificial Intelligence in Finance},
pages={144--155},
year={2023},
organization={SBC}
}