AI technology to identify foreign accounts suspected of attempting to influence public opinion in Korea that analyzes online news commentsKAIST
Scientists found some 24,000 foreign accounts, whose nationalities were undisclosed, that posted malicious comments on Korean news boards to fan internal animosity and conflict.
The findings were based on an AI-assisted analysis of 110 million comments posted between 2006 and 2025 on the news section of Naver, one of Korea’s major portal sites, by a joint research team led by KAIST professors Lee Won-jae, Cha Mee-young and Oh Hae-yeon and Thorsten Holz from the German Max Planck Institute.
The analysis started with 70 accounts previously flagged by the Institute for National Security Strategy for suspected links to foreign influence operations. Tracing their activity and follower connections uncovered another 23,998 suspicious accounts. The research team then developed an AI tool to analyze the suspicious 1.1 billion comments.
Most comments posted by the accounts sought to divide Korean society or portray politicians as incompetent and immoral, according to researchers. Typical examples included claims that “Korea’s military conscription is state-led exploitation” and comments attacking U.S. influence around the world.
Patterns observed over the 20 years showed that the accounts were particularly active during election seasons.
An average of 34.19 suspected foreign-linked accounts began operating each week during election periods,namelypresidential elections. This was about 51 percent more than the weekly average of 22.61 outside election periods.
Rather than posting comments favoring a particular political camp, however, the accounts attacked both progressives and conservatives. They appeared intent on fueling conflict across the political spectrum.
The researchers did not disclose where the users behind the accounts were from. Besides login locations, researchers identified suspected foreign users based on usernames and nonnative Korean expressions.
“AI allowed us to examine comments ranging from those that read like literal translations to those rendered more naturally by online translation tools,” Prof. Lee Won-jae of KAIST Graduate School of Culture Technology said.
The researchers then classified whether comments posted by the accounts contained emotions such as hatred, contempt or praise, and analyzed whether those emotions were directed at particular countries or targets. They combined those findings with factors that included how long the accounts had been active, how many comments they had posted and recurring behavioral patterns.
Activity of suspected foreign-linked accounts that attempted to influence public opinion in Korea between 2006 and 2025KAIST
The researchers called the technology “explainable AI” because it provides the reasoning behind its assessments, including the specific language that led it to flag a comment as suspected public opinion manipulation.
“Suspicious accounts consistently attacked Korea and its politicians and fueled divisions rather than targeting any particular political camp, a pattern we found across 20 years of data and successive administrations,” Lee said.
“This could help identify which messages warrant closer scrutiny during elections and other periods of heightened social tension.”
Prof. Cha also noted that the AI technology could become “a practical tool for improving transparency and trust in the digital public sphere.”
BY IM SOUNG-BIN [lee.soojung1@joongang.co.kr]
This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
