Isabelbinu/roberta-large-argumentscheme-fallacy-classifier

ArgueBot — Argument Scheme & Fallacy Classifier

This model is a fine-tuned version of roberta-large trained on a combined dataset of argument schemes and logical fallacies. It classifies any input text into one of 24 categories — either a valid argument scheme type or a logical fallacy type — in a single inference pass.

Model Details

Training Parameters

These are the exact parameters used to train this model:

Early Stopping

The best model checkpoint is saved automatically whenever validation loss improves. When early stopping triggers, the best weights are restored before evaluation.

Data Preprocessing

Supported Labels

✅ Argument Schemes (valid arguments)

1. 1.Argument from Analogy
2. 2.Argument from Alternatives
3. 3.Argument from Cause to Effect
4. 4.Argument from Commitment
5. 5.Argument from Example
6. 6.Argument from Expert Opinion
7. 7.Argument from Negative Consequences
8. 8.Argument from Positive Consequences
9. 9.Argument from Practical Reasoning
10. 10.Argument from Sign
11. 11.Argument from Values

⚡ Fallacy Types

1. 1.Ad Hominem
2. 2.Ad Populum
3. 3.Appeal to Emotion
4. 4.Circular Reasoning
5. 5.Equivocation
6. 6.Fallacy of Credibility
7. 7.Fallacy of Extension
8. 8.Fallacy of Logic
9. 9.Fallacy of Relevance
10. 10.False Causality
11. 11.False Dilemma
12. 12.Faulty Generalization
13. 13.Intentional

Dataset

• —Text column: Argument
• —Label column: Label
• —Both datasets combined into a single CSV before training

Results

How to Load and Use the Model

Step 1 — Install

pip install transformers torch

Step 2 — Load from Hugging Face

import json, torch, requests
from transformers import AutoTokenizer, AutoModelForSequenceClassification

MODEL_PATH = "Isabelbinu/roberta-large-argumentscheme-fallacy-classifier"

tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model     = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
model.eval()

meta      = requests.get(f"https://huggingface.co/{MODEL_PATH}/resolve/main/metadata.json").json()
label_map = {int(k): v for k, v in meta["label_map"].items()}
scheme_ids = set(meta["scheme_ids"])

def predict(text):
    inputs  = tokenizer(text, return_tensors="pt", truncation=True, max_length=128)
    with torch.no_grad():
        probs = torch.softmax(model(**inputs).logits, dim=-1).squeeze()
    pred_id = int(probs.argmax())
    return {
        "verdict":    "✅ Valid Argument" if pred_id in scheme_ids else "⚡ Fallacy",
        "label":      label_map[pred_id],
        "confidence": f"{probs[pred_id]:.1%}",
    }

# Test 1 
print(predict("Introducing a four-day work week will boost employee wellbeing, reduce burnout, and ultimately increase overall productivity."))


# Test 2
print(predict("Don't trust him — he was caught lying before, so everything he says is wrong."))

Applications

• —Education: Teach critical thinking by identifying argument types in real texts
• —Debate Analysis: Evaluate the quality of reasoning in speeches and essays
• —Fact-checking: Flag logically flawed reasoning in news and social media
• —Media Literacy: Help readers identify manipulation tactics in persuasive content
• —AI Assistants: Add argumentation reasoning to conversational AI systems

License

This model is licensed under the MIT License.