2026-10-09T18:05:42Z https://api.figshare.com/v2/oai

oai:figshare.com:article/32640459 2026-10-01T16:30:59Z portal_1172 item_type_8 month_year_10_2026

Evidence-based approach to verification of online health-related content Pritam Deka (24169275) PUREID: 617912715 Natural Language Processing AI fact verification health misinformation The prevalence of false information in online health articles, particularly highlighted during the COVID-19 pandemic, poses significant risks as people increasingly seek health-related advice online. While advances in machine learning (ML) and natural language processing (NLP) offer potential tools to help identify false health information, existing research has largely focused on political news. Health misinformation requires distinct approaches due to its reliance on current, reliable medical resources, which aren’t easily accessible in traditional fact-checking systems. This thesis addresses the verification of online health information using evidence-based medicine (EBM), emphasizing scientific rigor, reliable sources, and up-to-date knowledge. Given the scarcity of labeled data for training, the study explores unsupervised methods and transfer learning techniques, comparing these to existing supervised approaches. The thesis proposes a novel unsupervised method for detecting health claims and retrieving relevant scholarly resources. Additionally, a method is introduced to extract evidence sentences from credible medical sources, with evaluations against state-of-the-art baselines. The work further develops an approach to combine evidence sentences to assess claim veracity, while also exploring data augmentation techniques to address data limitations. These techniques vary across methods to tackle different challenges in data scarcity.<br><br> 2026-10-01T16:30:59Z Text Thesis 10.17034/32640459.v1 https://figshare.com/articles/thesis/Evidence-based_approach_to_verification_of_online_health-related_content/32640459 All Rights Reserved