Corresponding Author : Anamika Gupta

Co-Authors : Ayushi Gupta, Veenu Bhasin

Abstract :

Cyberbullying involves aggressive or harmful behavior conducted by individuals or groups through online platforms. Bystanders, those who witness such incidents, can play varying roles: defenders who intervene, instigators who support the bully, or neutral onlookers who remain passive. Understanding bystander behavior is crucial to assessing the scale and impact of cyberbullying, but the lack of labeled data limits progress. To address this, the CYBY23 dataset was released on Kaggle in October 2023, featuring Twitter threads with main tweets and bystander replies where bystander roles have been manually annotated. Labeling of bystander roles is a labor-intensive task; there's a clear need for automated methods to label bystander roles. In this paper, an efficient model, based on neural network (NN) architectures, is proposed for the automatic labeling of bystander role. CYBY23 dataset was first balanced using SMOTE. We then evaluated different sets of features (Toxicity, Sentiment, Empath, TF-IDF, POS) using three NN models - Feed-forward Neural Network, Extreme Learning Machine, and Multi-Layer Perceptron. The efficiency of the models was compared based on accuracy, precision, recall, and F1 score. Features TF-IDF and POS, when taken together, delivered the best performance. The feed-forward neural network gave the best performance among the network architectures.

Keywords :

Bystanders, Cyberbullying, Defender, Instigator, Impartial

Article Details

Unique Paper ID: 61003

Publication Volume & Issue: VOLUME 6 - 2026, ISSUE 1

Page(s): 30-43

Article Preview & Download

Share This Article

• X