Skip to content
Book Open access

ClusterRec: Cluster-Aware Recommendation with Large Language Models for Efficient and Scalable Inference

Nov 2026 · Proceedings of the 35th ACM International Conference on Information and Knowledge Management · 0 citations · 55 references

Abstract

Recent advances in large language models (LLMs) have significantly improved recommender systems by enabling richer semantic modeling of user preferences. However, existing LLM-based recommenders typically require expensive per-user inference over long interaction sequences, causing computational costs to scale rapidly with both model size and the number of users. To address this challenge, we propose ClusterRec, a framework that decouples LLM-based recommendation into two stages: (1) group-level preference modeling and (2) user-level personalization. Specifically, users are first clustered based on behavioral representations, and LLMs are employed to construct a group-level semantic representation for each cluster. Individual user representations are then constructed by integrating shared group representations with personalized collaborative filtering embeddings. This design amortizes LLM inference at the group level, substantially reducing computational overhead while preserving personalization. Extensive experiments on four real-world datasets demonstrate that ClusterRec achieves performance comparable to state-of-the-art methods, including conventional CF-SRec (e.g., SASRec) and recent LLM-based recommenders, while significantly reducing LLM inference cost. These results highlight that group-level semantic modeling provides an effective and scalable paradigm for integrating LLMs into practical recommender systems. The code is available at https://github.com/tsinghua-fib-lab/ClusterRec.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.