Scholarlink

Cyberbullying has become a severe real-time issue that significantly affects people in the digital era daily. Numerous studies are ongoing to identify and stop cyberbullying incidents on different internet platforms. Existing approaches primarily aim to identify cyberbullying content within a particular platform. This limitation served as the impetus for our research, which presents a unified framework for recognizing cyberbullying text on any social media platform. In this study, we provide an improved and novel unified, holistic approach to detect text cyberbullying, considering Facebook, Twitter, and YouTube platforms. Therefore, we aim to (i) develop a Source Detector Model (SDM) that discerns the source of the given text, whether it originates from Facebook, Twitter, or YouTube. (ii) develop a Platform Specific Model (PSM), the cyberbullying instances are then identified with the PSM. The fundamental highlight of our unified framework is its adaptability to various social media platforms, surpassing the constraints of single-source platform research. This paper considers Facebook, Twitter, and YouTube platform datasets, and it is trained with Machine Learning (ML) models to identify the cyberbullying content. Stochastic Gradient Descent (SGD) model achieved the highest accuracy scores of 0.966 for the Social Platform (SP) dataset and 0.801 for the Twitter dataset. Logistic Regression (LR) model demonstrated superior performance with an accuracy score of 0.801 for the Facebook dataset, while Random Forest (RF) model attained the highest score of 0.887 for the YouTube dataset. We attained significant improvements, making it a promising solution to combat cyberbullying across various social media platforms.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Empirical Analysis of Cyberbullying Detection for Social Networks with Machine Learning Pipeline—A Unified Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • A. Habiba,
  • G. Aghila

摘要

Cyberbullying has become a severe real-time issue that significantly affects people in the digital era daily. Numerous studies are ongoing to identify and stop cyberbullying incidents on different internet platforms. Existing approaches primarily aim to identify cyberbullying content within a particular platform. This limitation served as the impetus for our research, which presents a unified framework for recognizing cyberbullying text on any social media platform. In this study, we provide an improved and novel unified, holistic approach to detect text cyberbullying, considering Facebook, Twitter, and YouTube platforms. Therefore, we aim to (i) develop a Source Detector Model (SDM) that discerns the source of the given text, whether it originates from Facebook, Twitter, or YouTube. (ii) develop a Platform Specific Model (PSM), the cyberbullying instances are then identified with the PSM. The fundamental highlight of our unified framework is its adaptability to various social media platforms, surpassing the constraints of single-source platform research. This paper considers Facebook, Twitter, and YouTube platform datasets, and it is trained with Machine Learning (ML) models to identify the cyberbullying content. Stochastic Gradient Descent (SGD) model achieved the highest accuracy scores of 0.966 for the Social Platform (SP) dataset and 0.801 for the Twitter dataset. Logistic Regression (LR) model demonstrated superior performance with an accuracy score of 0.801 for the Facebook dataset, while Random Forest (RF) model attained the highest score of 0.887 for the YouTube dataset. We attained significant improvements, making it a promising solution to combat cyberbullying across various social media platforms.