finbert-full-sft-financial-sentiment_v3

FinBERT — Financial Sentiment (Fine-tuned on NOSIBLE)

Full fine-tuned ProsusAI/finbert for financial sentiment classification (positive / neutral / negative), trained on NOSIBLE Financial Sentiment (100K examples).

Model Details

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="poseidon1113/finbert-full-sft-financial-sentiment_v3"
)

classifier("Operating profit rose to EUR 13.1 mn from EUR 21.1 mn.")
# → [{'label': 'positive', 'score': 0.98}]

classifier("The company reported a loss for the third consecutive quarter.")
# → [{'label': 'negative', 'score': 0.95}]

Evaluation on FiQA 2018 (all splits combined)

Evaluated on pauri32/fiqa-2018 (train + validation + test combined) after removing URLs and non-ASCII characters.

Comparison across models

Why FinBERT outperforms GPT-2 LoRA

FinBERT's encoder-only BERT architecture reads the full sentence bidirectionally, making it inherently better at classification tasks. GPT-2 is a generative model adapted for classification via prompting — fundamentally less suited for this task. For production sentiment classification, encoder models like FinBERT are the right choice. GPT-2 LoRA is better suited for tasks requiring text generation.

Limitations

• Trained on English financial news only
• May underperform on social media slang or non-standard financial text
• Labels reflect financial impact sentiment, not general opinion sentiment

Citation

@article{araci2019finbert,
  title={FinBERT: Financial Sentiment Analysis with Pre-trained Language Models},
  author={Araci, Dogu},
  journal={arXiv preprint arXiv:1908.10063},
  year={2019}
}