Enterprise Organization of National Program of Excellence in Software

A paper from main Lab. (Advisor: Professor Jaekwang Kim) has been accepted for publication in the top-tier international conference, The 2026 ACM Web Conference (WWW) Research Track. The paper is scheduled to be presented in Dubai in April 2026. The paper titled "FCRLLM: Aligning LLM with Collaborative Filtering for Long-tailed Sequential Recommendation" involved Byung-moon Heo (PhD student, Artificial Intelligence Convergence Major), Nam-jun Lee (Master's student, Artificial Intelligence Convergence Major), and Seon-a Kim (Master's student, Department of Computer Science and Engineering) as authors, with Professor Jaekwang Kim participating as the corresponding author.

To address recommendation challenges for long-tailed users and items with insufficient interaction data, this research proposes the FCRLLM framework, which combines the rich semantic knowledge of Large Language Models (LLMs) with traditional collaborative filtering signals. The core technology, the Flipped Classroom mechanism, encourages dynamic alignment by allowing collaborative representations and semantic representations to interchangeably serve as teacher and student. During this process, a Hopfield Network-based energy function is used to minimize the differences in attention patterns between the two modalities, enabling complementary learning. The proposed method was evaluated through experiments on three real-world datasets, showing that it consistently improves recommendation performance regardless of item popularity or user activity levels. This study indicates that integrating multi-dimensional information through a bidirectional teacher-student structure allows for the development of more sophisticated and diverse recommendation systems.

| Professor Jaekwang Kim | linux@skku.edu

| main Lab. | sites.google.com/view/skku-milab