EMET: Embeddings from Multilingual-Encoder Transformer for Fake News Detection

Abstract

In the last few years, social media networks have changed human life experience and behavior as it has broken down communication barriers, allowing ordinary people to actively produce multimedia content on a massive scale. On this wise, the information dissemination in social media platforms becomes increasingly common. However, misinformation is propagated with the same facility and velocity as real news, though it can result in irreversible damage to an individual or society at large. Solving this problem is not a trivial task, considering the reduced size of the text messages usually posted on these communication vehicles. This paper proposes an end-to-end framework called EMET to classify the reliability of small messages posted on social media platforms. Our method leverages text-embeddings from multilingual-encoder transformers that take into consideration the semantic knowledge from preceding trustworthy news and the use of the reader’s reactions to detect misleading content. Our findings demonstrated the value of user interaction and prior information to check social media post’s credibility.