No abstract

While the virality of misinformation has been recognized as one of the significant global issues in the modern societies, few studies had examined the computational approaches to represent and identify false information in health domains. The current study aimed at using both psycholinguistic and natural language processing models to represent verified true and false texts about human papillomavirus (HPV) vaccines. Compared to the conventional word-embedding models representing texts in the levels of words, sentences or documents, results showed that introducing the embedding in the levels of propositions best differentiated the semantic representations in true and false texts. The study would advance our understandings in representing health texts and have implications on detecting false health information.