Citation

Ko, Yoon Thiri and Mohiuddin, Golam Md and Kimpan, Warangkhana (2026) Swarm-Inspired Ensemble Framework for Offensive Language Detection Using Transformer Models. Advances in Real-Time and Autonomous Systems, 1979. pp. 100-120. ISSN 2367-3370

Full text not available from this repository.

Abstract

The exponential growth of social media has amplified the spread of offensive and harmful language, presenting significant challenges for automated content moderation. Traditional detection systems often struggle with the contextual complexity and evolving nature of online discourse. This study proposes a swarm-inspired ensemble framework designed to enhance the adaptability and detection accuracy for offensive language. The framework integrates three high-performing transformer-based models, BERT, RoBERTa, and XLNet, via a weighted ensemble combination. Crucially, the optimal weights for this ensemble are determined using three distinct Swarm Intelligence (SI) algorithms: Particle Swarm Optimization (PSO), Whale Optimization Algorithm (WOA), and Moth-Flame Optimization (MFO). These optimizers navigate a three-dimensional search space to maximize the F1-score on the minority offensive class as . The results demonstrate that all three algorithms reliably achieved high performance, with the ensemble reaching maximum Accuracy of 0.8558 (WOA) and a peak of 0.7349 (MFO). The study finds that despite their different movement strategies, the optimizers converged to highly similar results. This convergence is attributed to the constrained, low-dimensional search space and the high correlation among the base models’ predictions, proving the efficacy and stability of SI techniques for automated weight tuning in highly constrained deep learning ensembles.

Item Type:	Article
Subjects:	P Language and Literature > P Philology. Linguistics
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	04 Sep 2026 01:33
Last Modified:	04 Sep 2026 01:38
URII:	http://shdl.mmu.edu.my/id/eprint/16671

Downloads

Edit (login required)