Detecting Urdu COVID-19 misinformation using transfer learning

Social media has become a prominent news source for many individuals due to its convenience and accessibility. Unfortunately, this accessibility has also facilitated the rapid dissemination of misinformation, irrespective of its credibility, particularly over the past decade. Misinformation spreaders take advantage of serious situations, like the COVID-19 pandemic, to spread false information. While numerous studies have focused on detecting misinformation in the English language, research in Urdu language misinformation detection remains limited. Our contribution to this field is twofold: firstly, we have compiled a substantial and diverse dataset of Urdu tweets. Secondly, we have introduced a novel approach incorporating feature extraction and ensemble learning techniques, complemented by high-performance filtering and voting classifiers designed explicitly for the COVID-19 Urdu dataset. The research paper employs a BERT-based transfer learning approach for feature extraction, followed by machine learning and ensemble methods for adequate classification. To evaluate the effectiveness of the proposed model, this paper conducted a comparative analysis against state-of-the-art methods. Experimental results demonstrate that the proposed model for detecting misinformation surpassed existing approaches when applied to the COVID-19 Twitter Urdu Tweets dataset, achieving an impressive F1 score of 0.93 and an accuracy of 0.92. This highlights the superior performance and efficacy of our method compared to previous methodologies.