Bridging the Gap: Self-Optimized Fine-Tuning for LLM-based Recommender Systems

2025-05-27

AI Summary

Large language models (LLMs) suffer from limited knowledge transfer capability in recommendation systems, as neither prompt-based guidance nor supervised fine-tuning alone can effectively bridge the gap between their general-purpose knowledge space and domain-specific recommendation requirements. To address this, we propose Self-Optimized Fine-Tuning (SOFT), a novel “guidance + fine-tuning” co-design paradigm. SOFT introduces self-distillation to construct semantically consistent, lightweight auxiliary recommendation data, and incorporates an adaptive curriculum scheduling mechanism to enable progressive learning—from easy to hard instances. Evaluated across multiple benchmarks, SOFT achieves an average 37.59% improvement in recommendation accuracy, significantly outperforming both pure prompting and pure fine-tuning baselines. The implementation is publicly available.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent years have witnessed extensive exploration of Large Language Models (LLMs) on the field of Recommender Systems (RS). There are currently two commonly used strategies to enable LLMs to have recommendation capabilities: 1) The "Guidance-Only" strategy uses in-context learning to exploit and amplify the inherent semantic understanding and item recommendation capabilities of LLMs; 2) The "Tuning-Only" strategy uses supervised fine-tuning (SFT) to fine-tune LLMs with the aim of fitting them to real recommendation data. However, neither of these strategies can effectively bridge the gap between the knowledge space of LLMs and recommendation, and their performance do not meet our expectations. To better enable LLMs to learn recommendation knowledge, we combine the advantages of the above two strategies and proposed a novel "Guidance+Tuning" method called Self-Optimized Fine-Tuning (SOFT), which adopts the idea of curriculum learning. It first employs self-distillation to construct an auxiliary easy-to-learn but meaningful dataset from a fine-tuned LLM. Then it further utilizes a self-adaptive curriculum scheduler to enable LLMs to gradually learn from simpler data (self-distilled data) to more challenging data (real RS data). Extensive experiments demonstrate that SOFT significantly enhances the recommendation accuracy (37.59% on average) of LLM-based methods. The code is available via https://anonymous.4open.science/r/Self-Optimized-Fine-Tuning-264E

Problem

• Bridging gap between LLM knowledge and recommendation systems
• Combining Guidance-Only and Tuning-Only strategies for better performance
• Enhancing recommendation accuracy via self-optimized fine-tuning (SOFT)

Innovation

• Combines Guidance and Tuning strategies
• Uses self-distillation for auxiliary dataset
• Adaptive curriculum scheduler for learning