This page was machine-translated and may differ from the original. View original

SKT's self-developed LLM recommendation model, "One Model 4.0," selected the top 4% of papers.

Google 우선 소스Published2026.01.26 11:31

"Thinking AI Recommendation Model" at the world's most prestigious AI conference

SK Telecom has received recognition for its research achievements in its self-developed AI recommendation model, achieving notable results at a world-renowned artificial intelligence conference.

SK Telecom (CEO Jae-heon Jeong) announced on the 25th that its research paper on the LLM (Large-Scale Language Model)-based AI recommendation model 'One Model' version 4.0 was selected as an 'Oral Session' subject corresponding to the top 4% at the global AI conference AAAI 2026.

AAAI (Association for the Advancement of AI) is considered one of the world's top AI conferences, along with NeurIPS, ICML, and ICLR.

This year, approximately 24,000 papers were submitted, the largest number ever, but the acceptance rate was only 18%, making competition even fiercer.

Among these, SKT's research was selected for the top 4% presentation session, which is considered a recognition of both technical perfection and practicality.

The paper published this time deals with the structure and performance improvement strategy of 'One Model 4.0' developed by SKT.

This model comprehensively understands various behavioral data such as customer clicks, usage history, and interests, and processes them in natural language, from selecting recommended products to explaining the reasons for recommendations to creating customized marketing messages for customers. It is a ‘thinking AI recommendation model.’

In particular, this version applies 'Reinforcement Learning for Reasoning', which has recently been attracting attention among global AI researchers.

This is a key technology that significantly increases the persuasiveness and accuracy of recommendations by allowing AI to improve its thinking skills through a process of generating and evaluating multiple answers on its own.

Here, we combined the collaborative filtering score utilized in existing recommendation systems to enhance the LLM-based inference ability and reflect the consistent user preferences simultaneously.

'One Model 4.0' has proven its effectiveness by going beyond research and applying it to actual services.

It is currently applied to SKT's rate plan recommendations, T World, T Membership, and T Direct Shop, and has achieved an effect of improving customer response, including click-through rates, by up to two times compared to the previous version.

SKT plans to expand its scope to various services, including T Universe, within this year.

SKT's 'One Model' has been steadily developing since the commercialization of version 1.0 in 2023.

The 1.0 study was accepted at CIKM 2023, and the 2.0 study, published in 2024, won the Best Paper Award at SIGIR 2024.

3.0 was then unveiled at the CIKM 2025 workshop and attracted the attention of global researchers.

This 4.0 AAAI 2026 announcement is another achievement that demonstrates that SKT's AI research capabilities are recognized globally.

Seok Ji-hwan, SKT AT/DT Data Manager, said, “One Model research has been consistently achieving results in international academic conferences. He said, “We are proving SKT’s AI technology prowess by collecting data,” and “We will continue to lead innovation in customer experience by conducting practical AI research that is directly connected to actual products and services.”
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.
배종인 기자
배종인 기자