finbert-finetuned-sentimentpulse

Fine-tuned ProsusAI/finbert for 3-class financial news sentiment classification (positive / negative / neutral).

Part of the SentimentPulse pipeline — an open-source replica of Bloomberg BQuant Textual Analytics.

Benchmark

Evaluated on 349 held-out examples from FinancialPhraseBank + FiQA-SA:

Model	Precision	Recall	F1
VADER	0.52	0.48	0.51
Loughran-McDonald	0.64	0.47	0.47
FinBERT zero-shot	0.79	0.75	0.76
FinBERT fine-tuned (this model)	0.92	0.92	0.92

Per-class (fine-tuned):

Class	Precision	Recall	F1	Support
negative	0.83	0.86	0.85	74
neutral	0.99	0.98	0.99	140
positive	0.90	0.90	0.90	135

Training

• Base model: ProsusAI/finbert
• Training data: FinancialPhraseBank (sentences_allagree, 2,264 sentences) + FiQA-SA (pauri32/fiqa-2018, 1,173 sentences) — 80/10/10 stratified split
• Epochs: 5 · Batch size: 16 · LR: 2e-5 · WeightedTrainer (inverse class frequency)
• Hardware: Apple MPS (~4.5 min)
• Framework: HuggingFace Transformers 4.41

Usage

from transformers import pipeline

pipe = pipeline(
    "text-classification",
    model="EomaxlSam/finbert-finetuned-sentimentpulse",
)

pipe("Apple beats Q1 earnings estimates by 12%")
# [{'label': 'positive', 'score': 0.94}]

pipe("Goldman Sachs fell sharply after issuing a guidance cut.")
# [{'label': 'negative', 'score': 0.91}]

Labels

ID	Label
0	negative
1	neutral
2	positive

Repository

github.com/Eomaxl/SentimentPulse