A KoBART-Based Model for Detecting Online Grooming Risk in Korean Conversations 


Vol. 51,  No. 3, pp. 562-573, Mar.  2026
10.7840/kics.2026.51.3.562


PDF Full-Text
  Abstract

As the online activities of children and adolescents have surged with the development of digital platforms, online grooming targeting them has become a serious social issue. Previous research on grooming detection has predominantly relied on English datasets and focused on analyzing fragmented features of individual messages, showing clear limitations—particularly in korean-language settings—in capturing the intelligent crime that gradually builds trust across the full conversational context. To overcome these limitations, This study proposes a context-based online grooming detection model optimized for the Korean language. To this end, we translated and refined the publicly available PAN12 dataset using DeepL and a large language model, then after final researcher verification, constructed a high-quality corpus for detection that reflects the informal, conversation-style use of korean and its conversational context. Based on this dataset, we fine-tuned KoBART to implement a binary classifier that processes the entire conversation sequence as a single input, enabling effective modeling of contextual flow. Experimental results showed that the proposed model converged stably, achieving high performance with an accuracy of 99.18% and an F1-Score of 0.9918. Notably, a confusion matrix analysis confirmed that the False Negative error rate—failing to identify actual grooming conversations— converged to zero, demonstrating the system's high reliability and effectiveness. This study is significant in that it experimentally proves the effectiveness of a context-aware approach in the under-researched field of Korean grooming detection, providing a crucial foundation for future technological development.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

J. Kang, M. Jung, S. Kim, S. Lim, S. Jung, W. Heo, Y. Park, "A KoBART-Based Model for Detecting Online Grooming Risk in Korean Conversations," The Journal of Korean Institute of Communications and Information Sciences, vol. 51, no. 3, pp. 562-573, 2026. DOI: 10.7840/kics.2026.51.3.562.

[ACM Style]

Jun-kyu Kang, Min-su Jung, Seong-min Kim, Sung-hun Lim, Seok-ho Jung, Won-jun Heo, and Yoo-hyun Park. 2026. A KoBART-Based Model for Detecting Online Grooming Risk in Korean Conversations. The Journal of Korean Institute of Communications and Information Sciences, 51, 3, (2026), 562-573. DOI: 10.7840/kics.2026.51.3.562.

[KICS Style]

Jun-kyu Kang, Min-su Jung, Seong-min Kim, Sung-hun Lim, Seok-ho Jung, Won-jun Heo, Yoo-hyun Park, "A KoBART-Based Model for Detecting Online Grooming Risk in Korean Conversations," The Journal of Korean Institute of Communications and Information Sciences, vol. 51, no. 3, pp. 562-573, 3. 2026. (https://doi.org/10.7840/kics.2026.51.3.562)
Vol. 51, No. 3 Index

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)






pISSN : 1226 - 4717
eISSN : 2287 - 3880