Document Type

Article

Publication Date

5-25-2026

Publication Title

Proceedings of the International AAAI Conference on Web and Social Media

Volume

20

Issue

1

Pages

276–290

Publisher Name

AAAI Press

Publisher Location

Washington, D.C., USA

Abstract

Cyberbullying remains a critical societal issue, with LGBTQ+ individuals disproportionately affected. Although previous work proposed general cyberbullying detection models, LGBTQ+-targeted cyberbullying detection remains relatively unexplored. SpectrumNet, a novel transformer-based model introduced in this paper, goes beyond conventional cyberbullying detection by adding conversational context and identity-aware modeling. SpectrumNet freezes the RoBERTa backbone and adds three key components: a hierarchical attention network to capture linguistic nuance, a GRU-based encoder to better capture comment history, and a dynamic fusion module to effectively weigh contextual signals. To address dataset imbalance, we apply focal loss and weighted sampling. Trained on a large, annotated Instagram dataset, SpectrumNet effectively differentiates between non-bullying, general bullying, and LGBTQ+-targeted bullying. In particular, it achieves strong recall on targeted content and excels at detecting subtle forms of discrimination often missed in isolation but evident within threaded interactions.

Comments

Author Posting © The Author(s), 2026. This article is posted here by permission of AAAI Press for personal use and non-commercial redistribution. This article was published open access in Proceedings of the International AAAI Conference on Web and Social Media, Vol. 20, Iss. 1,  (May 2026), https://doi.org/10.1609/icwsm.v20i1.42638.

Download

Share

COinS