Faux-Hate is an NLP-based project focused on detecting fake and hate content in social media posts. This system leverages advanced language modeling techniques and efficient fine-tuning strategies to combat misinformation and online toxicity.
The project uses tweets written in Hinglish (Hindi language using the English script), which are translated to English using the IndicTrans2 tool. Two BERT models are fine-tuned with Low-Rank Adaptation (LoRA) to efficiently classify each tweet as:
- Fake or Not Fake
- Hateful or Not Hateful
LoRA enhances the model’s adaptability while reducing computational costs, making it ideal for real-world, low-resource applications.
- Source: Scraped tweets from public social media platforms.
- Language: Hinglish (Hindi in English script).
- Preprocessing:
- Translated Hinglish to English using IndicTrans2.
- Cleaned, tokenized, and labeled data for fake and hate content classification.
- Base Model: BERT (
bert-base-uncased) - Adaptation Method: LoRA (Low-Rank Adaptation)
- Task: Binary classification for both fake and hate detection.
- Libraries:
- Hugging Face Transformers
- PEFT (Parameter-Efficient Fine-Tuning)
- PyTorch / Transformers Trainer
- Python
- NLTK
- NLP (Natural Language Processing)
- BERT (Bidirectional Encoder Representations from Transformers)
- LoRA Adapters (Efficient fine-tuning)
- IndicTrans2 (Hinglish to English translation)
- Hugging Face Transformers and Datasets
git clone https://github.com/yourusername/faux-hate.git
cd faux-hate