---
language: en
license: mit
tags:
- climate-change
- fallacy-detection
- text-classification
- deberta-v2-xlarge
- climate
- logic
- NLP
- classification
repo_url: https://github.com/gaanaman/fallacy_deberta
base_model: microsoft/deberta-v2-xlarge
datasets:
- fzanartu/FLICCdataset
pipeline_tag: text-classification
model-index:
- name: deberta-v2-xlarge-climate-fallacy
  results:
  - task:
      type: text-classification
      name: Text Classification
    dataset:
      name: FLICC Dataset
      type: flicc
      split: test
    metrics:
    - type: accuracy
      value: 0.6875
      name: Accuracy
    - type: f1
      value: 0.6916
      name: F1 Score
    - type: precision
      value: 0.7295
      name: Precision
    - type: recall
      value: 0.6875
      name: Recall
---

# Climate Fallacy Detector (DeBERTa-v2-xlarge)

A fine-tuned **DeBERTa-v2-xlarge** model for detecting logical fallacies in climate misinformation.

[![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/gaanaman/fallacy_deberta) [![Hugging Face](https://img.shields.io/badge/🤗%20Model-deberta--v2--xlarge--climate--fallacy-blue)](https://huggingface.co/Gaanaman/deberta-v2-xlarge-climate-fallacy)

This model is a fine-tuned version of **[microsoft/deberta-v2-xlarge](https://huggingface.co/microsoft/deberta-v2-xlarge)** on the **FLICC taxonomy** dataset. It detects logical fallacies in climate misinformation claims (e.g., *Ad Hominem*, *False Equivalence*, *Fake Experts*).

It successfully validates the results of the paper *"Detecting Fallacies in Climate Misinformation: A Technocognitive Approach"* (Zanartu et al., 2024), achieving comparable performance on consumer hardware (Mac M-Series).

## Performance
- **Test F1 Score (weighted)/ (macro):** 0.69 / 0.68
- **Validation Accuracy:** 0.72
- **Test Precision:** 0.73
- **Test Recall:** 0.69

## How to Use

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "Gaanaman/deberta-v2-xlarge-climate-fallacy"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "The climate has changed before naturally, so humans aren't causing it now."

inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
    logits = model(**inputs).logits

predicted_class_id = logits.argmax().item()
print(model.config.id2label[predicted_class_id])
# Output: Slothful Induction (or similar)