Metaheuristic Feature Selection for BERT Embeddings in Toxic Comment Classification
International Journal of Drug Delivery Technology · 06.07.2026 · 16 (62s)
DOI: 10.25258/ijddt.16.62s.192 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 06.07.2026. Verfasst von Divya Singhal und 5 weiteren Autor:innen.
Abstract (Originalsprache)
Online platforms generate a large volume of user-generated content, which often includes toxic or hate comments. Automated detection of such instances has become an important problem in natural language processing. Recent trends in this area have shown a high dependence on the use of transformer-based models, which are known to generate highdimensional features. However, these embeddings may contain redundant features that increase computational complexity. This study proposes an optimization-driven framework for toxic comment classification using BERT embeddings combined with the Whale Optimization Algorithm (WOA) for feature optimization. BERT is first used to generate semantic embeddings for textual comments. WOA is then employed to identify an optimal subset of embedding features that improves model efficiency while preserving classification performance. The reduced feature set is evaluated using multiple machine learning classifiers. Experimental results on the Jigsaw Toxic Comment dataset demonstrate that the proposed approach reduces the feature space by approximately 24% while maintaining comparable classification performance.
Zitieren
APA
Singhal, D., Verma, A., Gupta, A. K., Kumar, V., Bhardwaj, S., & Tayal, M. (2026). Metaheuristic Feature Selection for BERT Embeddings in Toxic Comment Classification. International Journal of Drug Delivery Technology, 16(62s). https://doi.org/10.25258/ijddt.16.62s.192
Vancouver
Singhal D, Verma A, Gupta AK, Kumar V, Bhardwaj S, Tayal M. Metaheuristic Feature Selection for BERT Embeddings in Toxic Comment Classification. International Journal of Drug Delivery Technology. 2026;16(62s). doi:10.25258/ijddt.16.62s.192
BibTeX
@article{Singhal2026,
author = {Singhal, Divya and Verma, Ankit and Gupta, Amit Kumar and Kumar, Vipin and Bhardwaj, Shashank and Tayal, Mahima},
title = {Metaheuristic Feature Selection for BERT Embeddings in Toxic Comment Classification},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {62s},
doi = {10.25258/ijddt.16.62s.192}
}
EUVIMED