"Something's Off About the Tone": Analyzing 100 Million Naver Comments

A comprehensive analysis of approximately 100 million Naver news comments by a joint research team from KAIST and Germany has revealed that tens of thousands of accounts are suspected of foreign-linked public opinion manipulation. Circumstances suggesting that these accounts fueled social conflicts—targeting both progressive and conservative factions alike—were also uncovered.

This is a report from TJB reporter Cho Hyeongjun.



Naver, South Korea's largest news portal, is estimated to receive millions of comments every day.

A joint international research team from KAIST and the Max Planck Institute in Germany conducted a comprehensive analysis of about 110 million comments written over a 20-year period from 2006 to last year.

The goal was to identify activities where specific countries or groups influence public opinion through online news comments.

The research team used an in-house developed artificial intelligence model, and the analysis method is as follows.

First, 70 foreign-linked accounts previously identified by the Institute for National Security Strategy served as the starting point.

Using AI, the team tracked accounts that were directly linked to these accounts or repeatedly wrote identical or similar comments on the same articles.

The primary targets were comments that used awkward Korean phrasing or could escalate social division, such as blind criticism or praise toward specific countries or figures.

As a result, out of 4.0 million Naver news users, approximately 24,000 accounts were identified as suspected foreign-linked influence operations.

[Lee Won-jae / Professor, KAIST Graduate School of Culture Technology: Applying 24 very conservative criteria, such as whether they posted the same number of comments at the same time, or whether they copied and pasted content....]

The characteristics of the comments written by these accounts also emerged; notably, there were far more contents of "criticism" than "praise."

It was also confirmed that these "critical" comments were broadly directed at various political forces, regardless of whether they were progressive or conservative.

In addition, during election periods characterized by intense political confrontation, such as presidential elections, the number of newly active suspicious malicious accounts increased by about 51% compared to usual.

[Lee Won-jae / Professor, KAIST Graduate School of Culture Technology: We were able to observe a phenomenon where peaks appear very consistently during such periods, whether it is a national election or a local election.]

The research team expects that these data analysis results will help detect organized and sophisticated online activities during periods of heightened social conflict.

(Photo: TJB / Choi Woon-gi, CG: TJB / Park Bo-young, Footprint courtesy of KAIST)

Cho Hyeongjun, TJB
※ Please note: This article was translated by AI and may contain errors.