Assessing User's Credibility to Enhance Deep Learning-Based Misinformation Detection on Social Media

The credibility of information on social media platforms are a necessity for ensuring that the content conveyed is trustworthy and exerts a positive influence on social media users. The spread of misinformation, such as rumors or fake news, is undesirable given that many people get the news and information first on their favorite social platforms and could be misled by such unauthentic information. One of the social media platforms that have been extensively used in such deceptive activities is Twitter. The automated detection of misinformation and preventing its spread on Twitter is a challenging task. One of the solutions is to determine the credibility of users and the content they generate through automated computing solutions. In this paper, we proposed a new mechanism to assess a user's credibility on Twitter to determine whether a user who shares a tweet is believable or trustworthy. This mechanism has been done by assigning different weights to the user's profile attributes (such as account creation date, Is the account verified or not, and account description) and social attributes (such as the number of followers and following) to compute the user credibility score. This score is then used to enhance the detection model of rumor and unverified information applicable to Twitter using the Bidirectional long-short-term memory (BiLSTM) approach. Experimental results showed that the model can detect rumors with an accuracy of 0.86.