Fake News Detection in Local Nigerian Languages (Hausa, Yoruba, Igbo, Pidgin)
# 📰 NaijFactCheck AI - Fake News Detection in Local Languages
## 🚀 Project Overview
This project aims to build an **AI & ML powered system** that detects fake news articles, social media posts, and online content in **local Nigerian languages** (Hausa, Yoruba, Igbo). The goal is to curb misinformation that spreads quickly in local communities and affects politics, health, and security.
## 📌 Problem
Fake news is a global challenge, but in Nigeria, the issue is amplified due to:
- Low digital literacy rates (UNESCO reports Nigeria has ~40% digital illiteracy).
- Widespread use of **local languages** in rural areas where misinformation spreads unchecked.
- Studies show that during elections, **over 50% of viral WhatsApp messages were misleading**.
- COVID-19 fake news in Hausa led to vaccine resistance in Northern Nigeria.
## 💡 Solution
We propose a **multilingual fake news detection system** that:
- Collects and processes text in **Hausa, Yoruba, and Igbo**.
- Uses **NLP (Natural Language Processing)** with **machine learning models** to classify news as **real or fake**.
- Provides a simple web/app interface for users to test the credibility of texts.
- Can be extended to WhatsApp/Facebook monitoring in the future.
## ⚙️ Tech Stack
- **Python** (NLTK, Hugging Face Transformers, scikit-learn, TensorFlow/PyTorch)
- **Data**: Web scraped Nigerian news (BBC Hausa, Premium Times, Punch, etc.)
- **Deployment**: Streamlit / Flask for demo
- **Languages**: Hausa, Yoruba, Igbo + English baseline
## 📊 Challenges
- Lack of large labeled datasets in local languages.
- Translating and pre-processing mixed-language (code-switching) text.
- Building lightweight models that can run on low-resource devices.
## 🛠️ Progress
- [ ] Data collection (news articles & social media posts)
- [ ] Text preprocessing & language translation
- [ ] Model training (Baseline: Logistic Regression, Advanced: BERT-based models)
- [ ] Evaluation & demo app
## 👥 Team
Currently: **1 member (open to c …