epistemic-stance-classifier

Epistemic Stance Classifier

A Longformer-based classifier for detecting epistemic stances (absolutist, evaluativist, multiplist) in text.

Model Details

• Model Type: LongformerForSequenceClassification (Longformer architecture)
• Base Model: allenai/longformer-base-4096
• Max Sequence Length: 2048
• Number of Labels: 3
• Labels: absolutist, evaluativist, multiplist

Training Details

• Best Validation Metric (f1_macro): 0.7081
• Best Epoch: 6
• Training Epochs: 6
• Learning Rate: 2e-05
• Batch Size: 4
• Gradient Accumulation Steps: 4
• Focal Loss: True
• Class Weights: True
• Temperature Scaling: True

Test Set Performance

• Accuracy: 0.7651
• F1 Macro: 0.6272
• F1 Weighted: 0.7574
• Expected Calibration Error: 0.0943

Usage

from transformers import AutoTokenizer, LongformerForSequenceClassification
import torch

model_name = "johnclund/epistemic-stance-longformer"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = LongformerForSequenceClassification.from_pretrained(model_name)

text = "Your text here..."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=2048)

with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class = torch.argmax(probs, dim=-1).item()

label_map = {0: "absolutist", 1: "evaluativist", 2: "multiplist"}
print(f"Predicted: {label_map[predicted_class]}")
print(f"Confidence: {probs[0][predicted_class]:.4f}")

Citation

If you use this model, please cite:

@misc{epistemic-stance-classifier,
  title={Epistemic Stance Classifier},
  author={John Lund},
  year={2026},
  howpublished={\url{https://huggingface.co/johnclund/epistemic-stance-classifier}}
}