EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Deep Linguistic Representation for High-Precision Cyberbullying Detection Using APO-Bi-LSTM with Attention Mechanism

Sathea Sree, Nalini Joseph L

International Research Journal of Multidisciplinary Technovation · 16.05.2026 · 236-256

DOI: 10.54392/irjmt26315 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Research Journal of Multidisciplinary Technovation, online erschienen am 16.05.2026. Verfasst von Sathea Sree und 1 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The phenomenon of cyberbullying on different social media has become severe following the further development of user-generated content, which calls for mechanisms that can detect expressions that are detrimental with high precision. Most of the current methods are fairly effective in dealing with offensive language, but are frequently unable to capture contextual subtleties, emotional polarities, and the constant shift in internet vocabulary. The goal of this research is to obtain a high-precision cyberbullying detector by systematically synthesizing an Adaptive Deep Linguistic Representation Model with a multimodal early-fusion strategy. This is achieved through an optimized framework utilizing a Bidirectional Long Short-Term Memory network and Attention Mechanism, tuned dynamically by Artificial Protozoa Optimization (APO-Bi-LSTM-AM). Appropriate publicly available social media text corpora with bullying, abusive, and neutral data is collected and narrowed down to clean data using cleaning tasks, such as noise elimination, lexical normalization, and contextual filtering of tokens. Internal semantic-refinement process is added to remove pretentiousness that emojis, irregular spellings, and redundant characters cause. To extract features, rich semantic vectors that can be used in downstream tasks of deep-learner are obtained using Word2Vec and contextual encoder-based representations. The APO-Bi-LSTM-AM combines both the bidirectional sequence learning and a focused attention layer, whereas the APO component regards some important network parameters that will promote convergence and minimize misclassification in borderline cases. The combination of these allows the architecture to obtain subtle cues, relational dependencies, and intensity variations in aggressive language, and is implemented in Python. Experimental assessments on the 600-instance test set indicate high levels of performance, identifying exactly 120 true positives and 468 true negatives. The model achieved a precision of 96.0%, accuracy of 98.0%, recall of 94.5%, and a corrected F1-score of 95.2%, outperforming traditional models. Its findings suggest that the Adaptive Deep Linguistic Representation Model suggested provides a powerful model of high-precision CBD in the social media setting.

Zitieren

APA

Sree, S., & L, N. J. (2026). Adaptive Deep Linguistic Representation for High-Precision Cyberbullying Detection Using APO-Bi-LSTM with Attention Mechanism. International Research Journal of Multidisciplinary Technovation, 236-256. https://doi.org/10.54392/irjmt26315

Vancouver

Sree S, L NJ. Adaptive Deep Linguistic Representation for High-Precision Cyberbullying Detection Using APO-Bi-LSTM with Attention Mechanism. International Research Journal of Multidisciplinary Technovation. 2026:236-256. doi:10.54392/irjmt26315

BibTeX
@article{Sree2026,
  author = {Sree, Sathea and L, Nalini Joseph},
  title = {Adaptive Deep Linguistic Representation for High-Precision Cyberbullying Detection Using APO-Bi-LSTM with Attention Mechanism},
  journal = {International Research Journal of Multidisciplinary Technovation},
  year = {2026},
  pages = {236--256},
  doi = {10.54392/irjmt26315}
}
Kostenlos starten →