Detecting Cyberbullying Among Young Students on Social Media in China: An Exploration Using Topic Modeling and NLP Techniques

• In Event: Supporting Youth Wellbeing in East Asia

Proposal

The digital age has dramatically altered the social landscapes navigated by young students. As social media platforms become deeply integrated into their lives, concerns about cyberbullying and its detrimental effects on mental well-being are rising. Specifically, in China, with its unique digital ecosystem and socio-cultural context, the ways in which cyberbullying manifests and its impact on the youth might differ from global patterns. This distinction underscores the need for dedicated research focused on Chinese social media platforms.

The primary objective of this research is to identify and dissect patterns of cyberbullying targeting young students on social media platforms prevalent in China. Rooted in the Social Cognitive Theory (SCT), this study acknowledges the pivotal role of observation in shaping behavior, especially in digital contexts. By observing negative behaviors online, individuals might emulate or even amplify these actions, leading to a surge in cyberbullying. The purpose of this study can be subdivided into three key goals: to craft a model proficient in detecting potential cyberbullying content; to understand the frequency, modality, and gravity of these occurrences; and lastly, to assess the sentiment that permeates discussions indicative of bullying.

A mixed-methods research design, incorporating both quantitative and qualitative data, is chosen for its adaptability and depth. This study takes Xiaohongshu, a social media platform that has gained immense popularity among young people in China in recent years, as a source of data. Data will be obtained from the social media platform's application program interface (API), focusing on content shared by or targeting young students. The robustness of this study depends on careful data selection. Public posts and comments are recognized sites of cyberbullying and are the focus of the study. The student user population further narrows the scope of the study and ensures the relevance of the survey. A sample of 200,000 random posts from this population over the past year will be analyzed to screen for posts involving cyberbullying as well as comments.

Text analytics tools are deployed for data analysis. First, Natural Language Processing (NLP) will refine the data and identify potential bullying cues. Then, topic modeling aims to reveal the major themes associated with cyberbullying incidents. In order to perform in-depth theme extraction, two well-known models are used, Latent Drichler Assignment (LDA) and Latent Semantic Analysis (LSA). In addition, to complement this analysis, sentiment analysis will measure the emotional tone of the suspected bullying discussion, whether it is negative, neutral or positive.

Reflecting on the broader implications, this research holds the promise of substantial impact. By elucidating the cyberbullying dynamics on popular social media platforms in China, the findings can guide educators, platform regulators, and guardians. Most crucially, by demystifying cyberbullying's manifestations, the research paves the way for timely interventions, ensuring a more secure digital environment for China's younger generation.

Author

• Jia Pang, Claremont Graduate University