Naver AI CleanBot 3.0 / Courtesy of Naver

Naver, Korea’s dominant internet portal, said Wednesday that it has upgraded its artificial intelligence (AI) moderation system to more aggressively filter content related to self-harm and the harassment of crime victims.

The update, known as “AI Cleanbot 3.0,” expands the portal’s automated oversight to include comments that glorify death or suicide. Introduced in November 2019 as the Korean internet industry’s first AI-based comment filter, the system will also target hateful language aimed at victims of accidents and crimes, as well as their families, amid rising concern over secondary victimization in online spaces.

The new version adds two priority detection categories for comments that promote or trivialize suicide, death or bodily harm, and derogatory, mocking or hateful comments directed at accident or crime victims and their families.

A key technical advance in version 3.0 is context-aware detection. Previous versions analyzed comment text in isolation; the upgraded model incorporates article headlines and body text to assess comments in their editorial context, allowing it to more accurately identify malicious intent even when offensive language is absent or disguised.

Naver said the system has evolved significantly since its initial release, which relied on keyword-based detection of profanity.

A 2020 upgrade extended detection to full-sentence context, enabling the system to flag insulting expressions without explicit profanity. Subsequent versions added detection of sexually offensive content, symbol- and character-based circumvention tactics and hateful or discriminatory expressions, including newly coined slang. The 2023 version incorporated guidelines on hate speech from the Korea Internet Self-governance Organization.

Beyond the technical upgrade, Naver said it has also implemented improved policy measures this year, including disabling comment sections on politics and election articles and automatically deactivating comments on articles that exceed a threshold of flagged content.

"We will continue strengthening CleanBot's performance to detect not only profanity and slurs but also newly emerging expressions of hate, degradation and discrimination," said a Naver official.

This article was published with the assistance of generative AI and edited by The Korea Times.

Source: Korea Times News