Bengali Emotion Classification from Social Media Text Using Deep Learning and Transformer‑Based Models with Explainability
Journal of Engineering Research and Sciences · 2026-06 · 5 (6) · 39-55
DOI: 10.55708/js0506004 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Engineering Research and Sciences, 2026-06. Verfasst von Mujtahid Alam und 4 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben.
Abstract (Originalsprache)
Bengali emotion classification remains challenging due to limited annotated resources, informal social media language, and the lack of comprehensive evaluations of modern transformer architectures. This study presents a unified framework for six‑class Bengali emotion classification using a corpus of 5,401 manually annotated social media comments. We systematically compare recurrent neural networks, transformer‑based models, and hybrid architectures, and propose a soft‑voting ensemble that integrates complementary contextual representations. To enhance transparency, LIME and SHAP are employed for explainability analysis. Experimental results show that the proposed ensemble achieves 91.31% accuracy and a weighted F1‑score of 0.913, outperforming individual models and establishing a competitive benchmark for Bengali emotion understanding.
Zitieren
APA
Alam, M., Aonty, S. S., Mahmud, S. N., Swachha, N. R., & Wazih, A. T. (2026). Bengali Emotion Classification from Social Media Text Using Deep Learning and Transformer‑Based Models with Explainability. Journal of Engineering Research and Sciences, 5(6), 39-55. https://doi.org/10.55708/js0506004
Vancouver
Alam M, Aonty SS, Mahmud SN, Swachha NR, Wazih AT. Bengali Emotion Classification from Social Media Text Using Deep Learning and Transformer‑Based Models with Explainability. Journal of Engineering Research and Sciences. 2026;5(6):39-55. doi:10.55708/js0506004
BibTeX
@article{Alam2026,
author = {Alam, Mujtahid and Aonty, Shuhena Salam and Mahmud, Sha Newaz and Swachha, Nahid Riaz and Wazih, Ahmed Talal},
title = {Bengali Emotion Classification from Social Media Text Using Deep Learning and Transformer‑Based Models with Explainability},
journal = {Journal of Engineering Research and Sciences},
year = {2026},
volume = {5},
number = {6},
pages = {39--55},
doi = {10.55708/js0506004}
}
EUVIMED