Beyond Fake or Real: A Journalism-Informed Taxonomy for Fine-Grained Misinformation Detection in Hindi Social Media
This preprint presents a journalism-informed taxonomy for distinguishing types of misinformation in Hindi social media. It introduces HM-800, a re-annotation of 800 posts from the CONSTRAINT 2021 dataset, with documented decision rules and annotation rationales. On 75 held-out examples spanning fabricated, manipulated, and misleadingly recirculated old content, a revised MuRIL training setup achieved a macro-F1 of 0.6819, compared with 0.2202 for a majority-class baseline on the same examples. The remaining errors highlight cases where verifying a claim requires evidence beyond the text itself. This manuscript has not been peer-reviewed.
Authors
- Tanmay Vaswani
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24405909
- Primary Topic
- Misinformation and Its Impacts
- Type
- preprint