A hybrid rumor detection model derived from a comparative study of supervised approaches

Abstract

In the current age of social media, information spreads like wildfire. Unfortunately, this also means that misinformation or rumors can spread easily. The spread of this misinformation can have negative consequences for society. This is especially true in recent years due to growing engagement in social media platforms for news. Hence, to prevent the spread of rumors, rumor detection is necessary. Bangladesh has been no exception to the spread of misinformation, causing countless propaganda over the years. Although a significant amount of work has already been conducted regarding rumor detection in English, Bangla rumor detection is still in its infancy. For our research, we first compared several Machine Learning (ML) models and Deep Learning (DL) models for rumor detection using both Bangla and English datasets. Comparing and analyzing the results, we implemented an Ensemble ML model and finally our hybrid model, which is a combination of our best-performing ML model and DL model that outperformed all other baseline state-of-the-art models.