Naver Webtoon "AI Recommendations Alleviate Popularity Concentration... Also Increase Diversity"

Naver Webtoon announced on the 10th that its AI-based personalized recommendation feature has alleviated concentration issues, providing equal opportunities for creators.

Naver Webtoon released a report titled "AI Curator Auto! The Effect of Enhancing Recommendation Diversity," analyzing the impact of its self-developed AI recommendation technology, 'AI Curator,' on increasing diversity in content consumption. This study was conducted by a research team led by Professor Geonwoong Lee from Korea University Business School.

AI Curator is a content recommendation technology developed by Naver Webtoon based on deep learning and machine learning.

In June this year, Naver Webtoon changed the default order of content exposure upon entering the Naver Webtoon application from 'popularity order' to 'Auto!' sorting based on AI Curator. 'Auto!' prioritizes recommending works based on users' tastes and preferences regardless of general popularity.

According to the research team's analysis, the 'Auto!' recommendation reduced market concentration across various criteria. A lower index indicates more active competition among companies and less market concentration.

Furthermore, AI recommendations considering users' preferences significantly increased the likelihood of works being selected and consumed, helping the works generate more value. When recommending works with 'Auto!', the average number of clicks was 59.5% higher compared to the 'popularity order' sorting. Views increased by 67.5%, and purchases were 62.7% higher.

The click-through rate relative to exposure was 68.6% higher, the view rate relative to exposure was 89.4% higher, and the purchase rate relative to exposure was 83.3% higher, indicating that works recommended by 'Auto!' are more likely to lead to actual views and purchases.

Additionally, Naver Webtoon's AI recommendation technology effectively addresses the structural difficulty of discovering and consuming new works, providing more balanced growth opportunities for a wider range of works.

For new works serialized within one year, those exposed through 'Auto!' recommendations had on average 4.5 times more clicks than non-new works, 3.7 times more views, and 3.4 times more purchases.