Typologies and Future Signals of School Bullying Using Social Big Data Analysis

Background/Purpose: Digital native adolescents spend a typical day with peers in both offline and online environments. The unique characteristics of online environments, such as anonymity and accessibility, have resulted in more diverse and complex types of school bullying. Despite the growing complexity of various types of school bullying, and evidence for their new experience in the online and offline domains, there is a paucity of research examining the typologies of school bullying across these two domains. Social big data may present promising opportunities to understand the rapid change in the participants, characteristics, and awareness of school bullying. To address the limitations, this study identified latent types and future signals of school bullying for both traditional and cyber bullying using social big data.

Methods: This study analyzed 478,879 online text data mentioning terms related to “school bullying”. The data was collected between January 1, 2013 and December 31, 2017, on 279 online channels (e.g., Twitter, Internet blog, online cafe, etc.) by a web crawler. First, the typologies were identified using Latent Dirichlet Allocation (LDA), which is a topic modeling method used in big data analysis. Second, word association was computed to describe the interrelation between multiple types of school bullying. Last, the noteworthy future signals were explored using the following methods: Term Frequency-Inverse Document Frequency (TF-IDF), Degree of Visibility (DoV), Degree of Diffusion (DoD), Keyword Emergence Map (KEM), and Keyword Issue Map (KIM).

Results: Buzz words describing school bullying have surged in the last two years and were concentrated between June and December, on Wednesday, and during 5 pm to 1 am. LDA results showed that the number of latent topics varied from year to year (two topics in 2013 and 2017; three topics in 2016; and four topics in 2014 and 2015). Each year, the most topics were characterized as direct bullying, on/offline crossed bullying, and group bullying; few topics dealt with indirect, exclusive, and single bullying. Topics related to physical and relational bullying appeared predominantly and constantly during 2013-2015; topics related to sexual and cyber bullying became increasingly apparent over time. As a result of word association analysis, words presenting verbal, relational, and sexual bullying behavior were concurrently used for both online and offline domains. The particular words, which represented stalking and infringement of personal information, were more intercorrelated with online than offline. TF-IDF results showed that the terms related to sexual and cyber bullying appeared as major keywords in recent years. The terms associated with digital devices (‘video’, ‘mobile phone’, ‘capture’) steadily emerged as marked keywords. Lastly, the current Strong Signals on school bullying were ‘cyber bullying’, ‘sexual harassment’, ‘group violence’, and the Future Signals were detected as ‘lynch’, ‘public shaming’, ‘doxing’, ‘faux arrest’.

Conclusions/Implications: By considering traditional offline bullying and online cyber bullying together on social big data, this research describes the present state of school bullying in South Korea. The typologies and future signals may contribute to understand experiences of digital native adolescents, and to develop further insights and comprehensive countermeasures for forthcoming school bullying problem.