---
base_model: microsoft/deberta-v3-base
library_name: peft
pipeline_tag: text-classification
language:
- en
tags:
- misinformation-detection
- social-media
- fakett
- peft
- lora
- transformers
- base_model:adapter:microsoft/deberta-v3-base
---

# DeBERTa — Text-Only Misinformation Detection on FakeTT

**Authors:** Andrei-Gabriel Radu, Ciprian-Octavian Truică, Elena-Simona Apostol

**National University of Science and Technology POLITEHNICA Bucharest**

LoRA adapter fine-tuned from `microsoft/deberta-v3-base` for binary **text-only misinformation classification** on the FakeTT social-media video dataset.

This model accompanies the bachelor thesis *Misinformation Detection in Social Media Videos*.


## Results

| Dataset | Modality | Macro-F1 |
|---|---|---:|
| FakeTT | Text-only | 0.7776 |

## Model

- **Base model:** `microsoft/deberta-v3-base`
- **Task:** Binary misinformation classification
- **Modality:** Text-only
- **Fine-tuning:** LoRA / PEFT
- **Dataset:** FakeTT
- **Number of classes:** 2
- **Primary metric:** Macro-F1

## Training

- **LoRA rank (`r`):** 8
- **LoRA alpha:** 32
- **LoRA dropout:** 0.05
- **Target modules:** `key_proj`, `query_proj`, `value_proj`, `dense`
- **Bias:** none

## Usage

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

repo_id = "DS4AI-UPB/deberta-misinfo-lora"
base_model_id = "microsoft/deberta-v3-base"

tokenizer = AutoTokenizer.from_pretrained(repo_id)
base_model = AutoModelForSequenceClassification.from_pretrained(base_model_id, num_labels=2)
model = PeftModel.from_pretrained(base_model, repo_id).eval()

text = "Example social media video description."
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

with torch.no_grad():
    logits = model(**inputs).logits

print(logits.argmax(dim=-1).item())
```

> Use the class-to-label mapping from the original FakeTT training pipeline.

## Intended Use

Research and benchmarking of English-language text-only misinformation detection for social-media video content.

## Limitations

This is a classification model, not a factual verification system. It cannot inspect the associated video and can degrade under domain shift.

## Citation

```bibtex
@thesis{radu2026misinformation,
    author = {Radu, Andrei-Gabriel and Truică, Ciprian-Octavian and Apostol, Elena-Simona},
    title  = {Misinformation Detection in Social Media Videos},
    school = {National University of Science and Technology POLITEHNICA Bucharest},
    year   = {2026}
}
```