Health-Message-Lens-Myth-Fact-Dataset

Purpose of Dataset

This dataset pairs health-related statements — written in the Hindi/Hinglish register common on WhatsApp forwards in India — with a myth/fact label and a short correction. It targets a specific gap: most published health-misinformation datasets used to train moderation and fact-checking classifiers are built from English-language, Western-sourced content, which doesn't match how health misinformation actually circulates in India. WhatsApp has over 97% market penetration in India (400M+ users) and is the dominant channel for both news and health misinformation. A cross-country study of COVID-19 misinformation found health-related claims made up the largest single category of misinformation analyzed. Separately, a rural-India field study found that a majority of misinformation posts observed had already been fact-checked by reputable agencies, yet continued circulating — suggesting existing fact-checks aren't reaching the audiences most exposed to this content, which motivates embedding detection directly in-channel rather than relying on external fact-check links. The dataset spans four categories: nutrition, vaccines, chronic_disease, and general_wellness. Each row includes the statement, a myth/fact label, category, and a one-sentence correction suitable for user-facing display.