Deep Learning Based Reliable User Identification in Social Media During Crisis

The identification of the trustworthy sources and determining their reliability during information diffusion are the most challenging tasks in social media. A trustworthy user identification system can greatly help authorities, organizations, and the public to access a accurate and fast information during crises, enabling a more knowledgeable and coordinated response. The proposed user’s reliable score prediction approach identifies the trusted user on social media. The proposed system analyzes their user profile, taking into account a number of factors, including their follower count, the number of accounts they follow, user verification status, and user’s profile description across the social media platforms to determine a user’s reliability and trustworthiness. In behavior analysis, the system analyzes the user behavior by considering the past tweets of the user and the engagement of the user through analyzing the likes and retweets count on the tweets. In tweet-relevant analysis, deep learning models are trained to classify whether the disaster tweets are relevant or irrelevant where the Bidirectional Long Short Term Memory model performs well with 94.54% accuracy. The reliable score is calculated with the user profile, behavior and tweet relevancy, and then the K-means clustering is performed to cluster the score to identify whether the user is highly reliable or not.