Workshop Proceedings of the 17th International AAAI Conference on Web and Social Media
Workshop: TrueHealth 2023: Workshop on Combating Health Misinformation for Social WellbeingDOI: 10.36190/2023.46
Misinformation is a major concern on social media and the Internet, especially during the COVID-19 pandemic. To address this problem, we propose a BERT-based explainable system to identify COVID-19 related misinformation, with a focus on explainability and transparency. In this paper, we present our approach of implementing the LIME XAI framework to explain the model's prediction process, making it transparent to both model practitioners and users. Additionally, we have developed a news source credibility checker to provide further context and credibility ranking for users. Our system uses TF-IDF and transformer-based embeddings to extract summary sentences from source articles, which are used alongside the claim to train the BERT-based classification model. Our system achieves high accuracy in classifying COVID-related claims as true, false, or partially true while also providing explainability and transparency, which is critical in combating misinformation. Our system provides users with not only the final prediction but also the explanation of how the prediction was made. The LIME XAI framework helps users understand which parts of the input data contributed the most to the prediction. This provides valuable insights into the classification process and increases trust in the system. In addition, our news source credibility checker provides further context for users to assess the reliability of the source articles. Our approach not only contributes to the fight against COVID-19 misinformation but also serves as a template for explainable and transparent machine learning models in other domains. In summary, our system provides an effective and explainable solution for detecting COVID-19-related misinformation on social media and the Internet.