An Improved Ensemble Framework for Social Media Fake News Detection Using RoBERTa Embeddings and the Whale Optimization Algorithm
TL;DR
A hybrid ensemble learning framework enhanced by RoBERTa-based feature representation and optimized using the Whale Optimization Algorithm is introduced to effectively capture deep semantic patterns in textual data while improving classification performance through adaptive parameter tuning.
Abstract
The increasing prevalence of fake news on social media has raised serious concerns due to its impact on public perception and decision-making. In response, this study introduces a hybrid ensemble learning framework enhanced by RoBERTa-based feature representation and optimized using the Whale Optimization Algorithm (WOA). The proposed approach aims to effectively capture deep semantic patterns in textual data while improving classification performance through adaptive parameter tuning. RoBERTa is employed to generate high-quality textual embeddings from news content, which are then fed into a set of base classifiers to ensure robustness and diversity in predictions. The WOA algorithm fine-tunes the ensemble model parameters, resulting in improved convergence and reduced error rates. The model was evaluated using two well-known fake news datasets, LIAR and ISOT. On the LIAR dataset, the proposed method achieved an accuracy of 98.17%, precision of 98.1%, recall of 97.8%, and an F1-score of 97.9%. On the ISOT dataset, it achieved an accuracy of 99.24%, precision of 99.3%, recall of 98.9%, and an F1-score of 99.1%. These results confirm the high reliability and balanced performance of the framework in distinguishing between true and false information across diverse content.