You can deploy this model and start earning money today!

syahrezapratama/
finbert-crypto-stance

Name: finbert-crypto-stance
Availability: InStock

safetensors

bert

Views

No views yet

An NLI-Based Approach to Asset-Specific Stance Detection in Cryptocurrency Tweets

This model classifies the stance of tweets toward Bitcoin (BTC) and Ethereum (ETH) as Bullish, Bearish, or Neutral using standard 3-class classification.

It was fine-tuned from ProsusAI/finbert as part of a master's thesis on NLI-based cryptocurrency stance detection.

How it works

Unlike the NLI models in this project, FinBERT uses direct 3-class classification. The existing sentiment head (positive/negative/neutral) is fine-tuned for stance detection (Bullish/Bearish/Neutral). Input is formatted as "{target} [SEP] {preprocessed_text}" to make the model target-aware.

Usage

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "syahrezapratama/finbert-crypto-stance"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
model.eval()

tweet = "BTC [SEP] Bitcoin is going to the moon! $100k is just the beginning"
inputs = tokenizer(tweet, return_tensors="pt", truncation=True, max_length=128)

with torch.no_grad():
    logits = model(**inputs).logits
    probs = torch.softmax(logits, dim=-1)

labels = {0: "Bearish", 1: "Neutral", 2: "Bullish"}
predicted = labels[torch.argmax(probs, dim=-1).item()]
print(f"Predicted stance: {predicted}")

Performance

Evaluated on a held-out test set of 450 tweets (70/15/15 train/val/test split, seed=42).

Overall metrics

Metric	Value
Accuracy	71.78%
Macro F1	0.6730
Weighted F1	0.7230

Per-class metrics

Class	Precision	Recall	F1	Support
Bearish	0.6042	0.6170	0.6105	47
Neutral	0.8374	0.7574	0.7954	272
Bullish	0.5641	0.6718	0.6132	131

Comparison with baselines

Model	Paradigm	Accuracy	Macro F1
FinBERT	Zero-Shot	58.44%	0.3543
FinBERT	Fine-Tuned	71.78%	0.6730
GPT-4o	Zero-Shot	76.67%	0.7275

Training details

Dataset

Source: 3,000 cryptocurrency tweets about BTC and ETH
Labels: Bullish (873), Neutral (1,814), Bearish (313)
Split: 2,099 train / 451 val / 450 test (seed=42)

Hyperparameters

Parameter	Value
Base model	ProsusAI/finbert
Paradigm	3-class classification
Learning rate	2e-5
Batch size	16
Max epochs	5 (early stopping patience = 3)
Max sequence length	128
Warmup	10% linear warmup + linear decay
Weight decay	0.01
Class weights	Bearish=3.19, Neutral=0.55, Bullish=1.15
Optimizer	AdamW
Best epoch	Early stopped based on validation macro F1

Limitations

Domain-specific: Trained only on cryptocurrency tweets (BTC and ETH). May not generalize to other financial assets or domains.
Class imbalance: Bearish tweets are underrepresented (10.4% of data), leading to lower recall on bearish stance despite class weighting.
Language: English only.
Temporal: Trained on tweets from a specific time period. Cryptocurrency language and sentiment patterns evolve rapidly.

Citation

If you use this model, please cite:

bibtex

@mastersthesis{pratama2026cryptostancenli,
  title={An NLI-Based Approach to Asset-Specific Stance Detection in Cryptocurrency Tweets},
  author={Pratama, Syahreza},
  year={2026},
}

License

MIT