I-S2FND: a novel interpretable self-ensembled semi-supervised model based on transformers for fake news detection

One of the serious consequences of social media usage is fake information dissemination that locomotes society towards negativity. Existing solutions focus on supervised fake news detection models, which requires extensive labelled data. In this paper, we deal with two different problems of fake news detection such as (1) Detecting fake news with limited annotated data and (2) Interpretability of the proposed model on fake news detection. We address these issues by designing an Interpretable Self Ensembled Semi-Supervised Fake News Detection Model (I-S 2 FND). In I-S 2 FND, the model learns the enhanced representations of labelled and unlabelled fake news by incorporating an adaptive pseudo-labelling mechanism on unlabelled data. Moreover, interpretation of the model on text using the gradients improves the identification of essential words in the content of fake news. Based on the experimental findings, it is evident that the proposed model outperforms existing state-of-the-art models by approximately 5% in terms of accuracy when trained with only a limited amount of labeled data across different datasets.