Horizon-Truth/horizon-truth-misinformation-dataset

A curated Ethiopian-context dataset of misinformation claims designed for fact-checking, research, and AI-based misinformation detection. # Horizon Truth - Misinformation Detection Dataset ## Overview This repository contains the dataset development resources for the **Horizon Truth Misinformation Detection System**. Horizon Truth is an AI-supported platform designed to help identify and understand harmful misinformation by combining automated analysis, verification approaches, and community-driven reporting. The project focuses on improving information reliability and digital safety, especially in local contexts. ## Dataset Status **Current Version:** Version 1 (Development Phase) **Status:** Under development This dataset is currently private and is not ready for public release. The repository is used for: - Dataset collection - Data cleaning - Annotation management - Version tracking - Model development preparation Future versions will expand coverage, categories, and labeling capabilities. --- # Version 1 Scope Version 1 focuses on two high-impact misinformation categories: | Category | Local Name | Purpose | |---|---|---| | Scams | ማጭበርበር (Finya) | Protect users from financial fraud and deception | | Health | ጤና (Fayyaa) | Reduce harmful health misinformation | --- # Veracity Labels The dataset uses a 4-class verification system: | Label | Class | Definition | |---|---|---| | 0 | TRUE | Verified by two or more credible sources | | 1 | FALSE | Fabricated or contradicted by reliable evidence | | 2 | MISLEADING | Partially true but missing context or presented deceptively | | 3 | UNVERIFIABLE | Not enough evidence to confirm or deny | ---