Automatic classification of participant roles in cyberbullying: Can we detect victims, bullies, and bystanders in social media text?

Abstract

Successful prevention of cyberbullying depends on the adequate detection of harmful messages. Given the impossibility of human moderation on the Social Web, intelligent systems are required to identify clues of cyberbullying automatically. Much work on cyberbullying detection focuses on detecting abusive language without analyzing the severity of the event nor the participants involved. Automatic analysis of participant roles in cyberbullying traces enables targeted bullying prevention strategies. In this paper, we aim to automatically detect different participant roles involved in textual cyberbullying traces, including bullies, victims, and bystanders. We describe the construction of two cyberbullying corpora (a Dutch and English corpus) that were both manually annotated with bullying types and participant roles and we perform a series of multiclass classification experiments to determine the feasibility of text-based cyberbullying participant role detection. The representative datasets present a data imbalance problem for which we investigate feature filtering and data resampling as skew mitigation techniques. We investigate the performance of feature-engineered single and ensemble classifier setups as well as transformer-based pretrained language models (PLMs). Cross-validation experiments revealed promising results for the detection of cyberbullying roles using PLM fine-tuning techniques, with the best classifier for English (RoBERTa) yielding a macro-averaged ${F_1}$ -score of 55.84%, and the best one for Dutch (RobBERT) yielding an ${F_1}$ -score of 56.73%. Experiment replication data and source code are available at https://osf.io/nb2r3.

Artificial Intelligence to Address Cyberbullying, Harassment and Abuse: New Directions in the Midst of Complexity

“…Many of the classifiers are still able to only categorise content or behaviours as abusive or not abusive, for instance, based on hateful language or slurs, and they do not provide a more nuanced description of the type of abuse involved, or the severity of the case; or roles played by those involved (for examples of multi-class classifiers; see, e.g. Balakrishnan et al, 2019;Jacobs et al, 2020). Furthermore, for a case of abuse to be classified as "cyberbullying", there still (according to widely used definitions) typically needs to be some level of repetition, intent to hurt and even power imbalance (although said definitions are under revision and tend to be intensely debated) 1 and classifiers for the most part do not capture those (Cheng et al, 2020).…”

A Systematic Review of Machine Learning Algorithms in Cyberbullying Detection: Future Directions and Challenges

“…Similarly, more perpetrators among [81] recognized bullying temporal patterns in the predator's questions using time series modeling methodology. Jacobs et al [161] tried to identify automatically different participants in a cyberbullying event from textual cyberbullying traces. [161] tried to identify automatically various participants in a cyberbullying event from textual cyberbullying traces.…”

“…Jacobs et al [161] tried to identify automatically different participants in a cyberbullying event from textual cyberbullying traces. [161] tried to identify automatically various participants in a cyberbullying event from textual cyberbullying traces. Although the F1-score of their best model is not very high (56.7%), there is a lot of scope in this area.…”

Implementation of Hyperparameter Optimisation and Over-Sampling in Detecting Cyberbullying Using Machine Learning Approach

“…By adopting features used in [28], [65], we extracted six categories of features called term lists, which are derived from one binary feature. All redundant terms were discarded and only 2950 unique terms were listed.…”