[Paper Review] Session-based Cyberbullying Detection in Social Media: A Survey

Peiling Yi, Arkaitz Zubiaga|arXiv (Cornell University)|Jul 14, 2022

Hate Speech and Cyberbullying Detection4 citations

TL;DR

This survey proposes a session-based cyberbullying detection framework that models repetitive behavior and power imbalance in social media interactions. It reviews data, methods, and best practices for dataset creation, benchmarks state-of-the-art models and large language models on two datasets, and identifies key challenges for future research in session-level cyberbullying detection.

ABSTRACT

Cyberbullying is a pervasive problem in online social media, where a bully abuses a victim through a social media session. By investigating cyberbullying perpetrated through social media sessions, recent research has looked into mining patterns and features for modeling and understanding the two defining characteristics of cyberbullying: repetitive behavior and power imbalance. In this survey paper, we define the Session-based Cyberbullying Detection framework that encapsulates the different steps and challenges of the problem. Based on this framework, we provide a comprehensive overview of session-based cyberbullying detection in social media, delving into existing efforts from a data and methodological perspective. Our review leads us to propose evidence-based criteria for a set of best practices to create session-based cyberbullying datasets. In addition, we perform benchmark experiments comparing the performance of state-of-the-art session-based cyberbullying detection models as well as large pre-trained language models across two different datasets. Through our review, we also put forth a set of open challenges as future research directions.

Motivation & Objective

• To define a standardized session-based cyberbullying detection framework that captures repetitive behavior and power imbalance in social media interactions.
• To review existing data sources and methodologies for session-based cyberbullying detection across diverse social media platforms.
• To establish evidence-based criteria for creating high-quality, representative session-based cyberbullying datasets.
• To benchmark state-of-the-art session-based models and large pre-trained language models on two public datasets.
• To identify open challenges and future research directions in session-based cyberbullying detection.

Proposed method

• The paper introduces a structured framework that decomposes session-based cyberbullying detection into data collection, feature engineering, modeling, and evaluation stages.
• It analyzes session-level features such as message sequence patterns, linguistic cues, and interaction dynamics to model repetitive behavior and power imbalance.
• The authors propose a set of best practices for dataset construction, including session segmentation, labeling consistency, and data balance across bully-victim dynamics.
• Benchmarking is conducted using state-of-the-art sequential models (e.g., RNNs, Transformers) and large pre-trained language models (e.g., BERT) on two public datasets.
• The evaluation uses standard NLP metrics such as F1-score, precision, and recall to compare model performance across different session configurations.
• The framework integrates insights from computational linguistics, social psychology, and NLP to ensure methodological and ethical robustness.

Experimental results

Research questions

• RQ1 How can a standardized framework be designed to model session-based cyberbullying using repetitive behavior and power imbalance?
• RQ2 What are the key data and methodological challenges in creating reliable session-based cyberbullying detection datasets?
• RQ3 How do state-of-the-art sequential models and large pre-trained language models perform on session-based cyberbullying detection tasks?
• RQ4 What evidence-based criteria can guide the creation of high-quality, representative session-based cyberbullying datasets?
• RQ5 What are the major open challenges and future research directions in session-based cyberbullying detection?

Key findings

• The proposed session-based detection framework effectively captures the temporal and relational dynamics of cyberbullying through structured session modeling.
• Benchmarking reveals that large pre-trained language models generally outperform traditional sequential models on session-based cyberbullying detection tasks.
• The study identifies significant performance variation across datasets, highlighting the need for standardized evaluation protocols.
• Evidence-based dataset creation criteria improve data quality, consistency, and reproducibility in future research.
• Key challenges remain in handling long sessions, detecting subtle power imbalances, and ensuring model generalization across diverse social media platforms.
• The survey underscores the importance of integrating social psychological insights with NLP techniques for more robust detection systems.