Naver enhances AI filtering for auto-complete search to combat rigged keywords

[Courtesy of Naver Corp.]

Naver will enhance its search filtering system using generative artificial intelligence (AI) technology, after observing frequent cases where intentional marketing keywords, such as local business names, appear in the auto-complete search results.

Auto-complete search is a service where related search terms, like “train tickets for Chuseok,” “Chuseok holiday,” and “Chuseok gifts,” automatically appear when users type keywords such as “Chuseok” in Naver’s search bar.

This feature helps provide a more efficient search environment by reflecting users’ interests.

However, problems arise when keywords like “Gangnam restaurant” or “Seoul dentist” are searched, leading to the auto-complete feature showing commercial results that have been introduced through irregular methods. This issue has reportedly resulted in a steady stream of user complaints toward Naver.

Naver announced that it will integrate a separate filtering process based on its large-scale language model (LLM), HyperCLOVA X, by the end of this month. This move aims to prevent low-quality keywords from appearing during the creation of auto-complete search terms.

“Auto-complete is a service designed to predict and suggest queries based on what users are likely to search for, helping them quickly find the information they need,” said a company official. “However, due to a recent increase in abnormal traffic from certain commercial keywords, we have decided to implement measures to improve the quality of the service.”

Naver plans to filter out auto-complete keywords that contain simple typographical errors, special characters, meaningless text or numbers, duplicated words, or terms that are irrelevant to the search query.