Towards Efficient Reasoning in LLM-Based Recommender Systems via Model Merging

2026-08-11

AI Summary

This work addresses the high computational cost in large language model–based recommender systems caused by lengthy reasoning in slow-thinking models, a challenge inadequately mitigated by existing compression methods that either incur substantial training overhead or suffer from poor scalability. The paper proposes the first fine-grained model fusion framework tailored for recommendation, which seamlessly integrates the behavior of fast-thinking models into slow-thinking models without additional training by selectively injecting representations within a shared parameter space. Crucially, it assigns head-specific fusion coefficients across attention heads to precisely preserve critical reasoning evidence while suppressing redundancy. This approach significantly shortens inference sequences—reducing length by up to 24.3% across three benchmark datasets—while maintaining or even improving recommendation accuracy, outperforming current fusion baselines.

Abstract

Large language model-based recommender systems are increasingly adopting slow-thinking models that generate step-by-step reasoning before making predictions, often achieving higher accuracy than fast-thinking models that predict directly. However, their reasoning traces are often unnecessarily verbose, increasing inference costs without commensurate accuracy gains. Existing training-based approaches to reasoning compression often incur substantial adaptation costs, while inference-time methods are brittle and difficult to scale. These limitations motivate model merging as a promising training-free direction for transferring specialised behaviours between models in a shared parameter space. In particular, merging a slow-thinking model with a fast-thinking counterpart provides a natural mechanism for balancing recommendation accuracy and reasoning conciseness. To this end, we propose, to our knowledge, the first model merging framework for reasoning compression in recommender systems. Unlike conventional merging methods that apply uniform merge coefficients across model components, our method performs fine-grained merging at the level of individual attention heads, capturing heterogeneous patterns in recommendation reasoning. Each attention head is assigned a distinct merge coefficient according to its contribution to critical reasoning evidence and its sensitivity to parameter change, enabling selective injection of the concise behaviour of the fast-thinking model into the slow-thinking model and reducing reasoning verbosity without compromising recommendation quality. Experiments on three benchmark datasets show that our method reduces reasoning length by up to 24.3% while outperforming competitive model merging baselines in maintaining recommendation accuracy. The code is available at https://github.com/linhledieu/REAM.

Problem

• reasoning compression
• LLM-based recommender systems
• model merging
• inference efficiency
• reasoning verbosity

Innovation

• model merging
• reasoning compression
• fine-grained merging
• attention heads
• LLM-based recommender systems