SEOUL, Jan 25: SK Telecom (NYSE: SKM) has been recognized for its research achievements in AI recommendation models that understand individual behavioral contexts and explain the reasons behind recommendations, earning a spot on the stage of a world-renowned AI conference.
At the 40th Annual AAAI Conference on Artificial Intelligence (AAAI-26) held in Singapore from January 20 to 27, 2026, the company presented its research paper “Think Wise, Collaborate Effectively: A Rationale-Aware LLM-Based Recommender with Reinforcement Learning from Collaboration Signals” as an oral presentation on January 24 in the session Data Mining & Knowledge Management Recommender Systems 4.
The AAAI Conference on Artificial Intelligence promotes theoretical and applied AI research as well as intellectual interchange among researchers and practitioners. It is regarded as one of the world’s top AI conferences, alongside the Conference on Neural Information Processing Systems (NeurIPS), the International Conference on Machine Learning (ICML), and the International Conference on Learning Representations (ICLR). This year, the number of paper submissions increased to around 24,000 from last year’s 13,000, while the acceptance rate dropped to 17.6% from 23%, making the competition more intense. Only around 4% of submissions were selected for oral presentation.
In addition to its participation in the oral session, SK Telecom also presented a poster on the same day during Poster Session: Data Mining & Knowledge Management 3.
The presented paper covers research on the company’s self-developed LLM-based AI recommendation model One Model Version 4. It also introduces methods for applying “reasoning ability reinforcement learning,” which has recently attracted attention among AI researchers worldwide, to real service environments.
One Model Version 4 comprehensively understands various behavioral data such as customer clicks, usage history, and interests, and generates in natural language which products or services to recommend; why those recommendations are made; and how to compose marketing messages for customers.
In particular, the model goes beyond simple data training by introducing “reasoning ability reinforcement learning,” enabling the AI to improve its own decision-making process and significantly enhance the persuasiveness and accuracy of its recommendations. “Reasoning ability reinforcement learning” refers to a training method in which AI develops its reasoning skills by generating multiple answers on its own and having them evaluated.
One Model Version 4 is also trained using collaborative filtering scores – used in traditional recommendation methods – for AI learning, enabling the model to reflect both LLM-based reasoning and consistent user preferences.
One Model Version 4 has been applied to a range of SK Telecom’s products and services, where its effectiveness in enhancing recommendation performance has been demonstrated in practice. For instance, it is currently being used in rate plan recommendations, T World (online customer support), T Membership (customer membership program), and T Direct Shop (official online store), achieving up to twice the customer response, such as click-through rates, compared to the previous version. Within this year, SK Telecom plans to expand the use of the model to various other products and services, including T Universe.
Since its initial commercial release in 2023 (Version 1), One Model has continuously improved recommendation performance and learning efficiency. Research on One Model Version 1 was accepted at the Conference on Information and Knowledge Management (CIKM) 2023, while the second version of the model received the Best Paper Honorable Mention at International ACM SIGIR Conference on Research and Development in Information Retrieval 2024. One Model Version 3 was introduced at the Conference on Information and Knowledge Management (CIKM) 2025 Workshop, where it garnered considerable attention.
“We have once again demonstrated our AI Capabilities by achieving meaningful recognition at world-renowned conferences with our One Model research,” said Seok Ji-hwan, Head of AT/DT Data Office. “We will continue to conduct practical AI research that can be directly applied to actual products and services.”