metadata
license: mit
datasets:
- nlp-brin-id/unsup-fact
language:
- id
Contrastive embedding trained unsupervisedly to implicitly contradict factual evidences and claims . The dataset is based on online public report for fake news detection. Model is trained based on contrastive loss and 'Fake news classification' as validation task. MLP is used as external classifier for validation task given the resulting embedding from trained model.
Code implementation will be shared publicly soon.