Implementing a Hybrid Parallel Framework Utilizing Machine and Deep Learning for Rapid Rumor Detection in Social Media

Abstract

Numerous businesses have faced significant repercussions due to the widespread dissemination of false information and rumors across social media platforms. The impact of fake news extends to tarnishing public perception, damaging corporate reputations, disrupting communities, undermining governmental integrity, exposing companies to risks, and posing a grave threat to social cohesion. This research article delves into the endeavours of prominent researchers focused on utilizing machine learning for rumour detection. Additionally, it explores a newly proposed framework wherein several established methods viz. Adaboost, Hard Voting, Gradient Boosting, and Random Forest; and a novel hybrid deep learning model CNN + BiLSTM + BiGRU operate simultaneously to identify rumours in a parallel environment. Utilizing time-series vector representations of Twitter, Facebook and FakeNewsNet datasets, this study suggests an ensemble approach for rumor detection. The proposed model demonstrates better accuracy, f1-score, recall, and efficiency compared to existing models and minimizes time consumption due to parallel computational capabilities.