Beyond Semantic Understanding: Preserving Collaborative Frequency Components in LLM-based Recommendation

2025-08-13

AI Summary

LLM-based recommender systems often over-rely on semantic similarity, causing attenuation of collaborative signals embedded in pre-trained ID representations during inference and degrading ranking performance. To address this, we propose a spectral-aware collaborative signal preservation framework grounded in spectral graph theory. First, we introduce Global Graph Low-Pass Filtering (G-LPF) to suppress noise in ID embeddings. Second, we design a Temporal Frequency Modulation (TFM) mechanism that dynamically amplifies collaborative-dominant low-frequency components across LLM layers. We establish, for the first time, a theoretical connection between local graph Fourier transforms and frequency-domain filtering, proving that TFM ensures layer-wise stability of collaborative signals. Evaluated on four benchmark datasets, our method achieves up to an 8.00% improvement in NDCG@10 over state-of-the-art approaches, demonstrating the critical role of spectral balance in enhancing the robustness of LLM-based recommendation.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Recommender Systems
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recommender systems in concert with Large Language Models (LLMs) present promising avenues for generating semantically-informed recommendations. However, LLM-based recommenders exhibit a tendency to overemphasize semantic correlations within users' interaction history. When taking pretrained collaborative ID embeddings as input, LLM-based recommenders progressively weaken the inherent collaborative signals as the embeddings propagate through LLM backbones layer by layer, as opposed to traditional Transformer-based sequential models in which collaborative signals are typically preserved or even enhanced for state-of-the-art performance. To address this limitation, we introduce FreLLM4Rec, an approach designed to balance semantic and collaborative information from a spectral perspective. Item embeddings that incorporate both semantic and collaborative information are first purified using a Global Graph Low-Pass Filter (G-LPF) to preliminarily remove irrelevant high-frequency noise. Temporal Frequency Modulation (TFM) then actively preserves collaborative signal layer by layer. Note that the collaborative preservation capability of TFM is theoretically guaranteed by establishing a connection between the optimal but hard-to-implement local graph fourier filters and the suboptimal yet computationally efficient frequency-domain filters. Extensive experiments on four benchmark datasets demonstrate that FreLLM4Rec successfully mitigates collaborative signal attenuation and achieves competitive performance, with improvements of up to 8.00% in NDCG@10 over the best baseline. Our findings provide insights into how LLMs process collaborative information and offer a principled approach for improving LLM-based recommendation systems.

Problem

• LLM-based recommenders overemphasize semantic correlations in user interactions
• Collaborative signals weaken as embeddings propagate through LLM layers
• Need to balance semantic and collaborative information for better recommendations

Innovation

• Global Graph Low-Pass Filter purifies item embeddings
• Temporal Frequency Modulation preserves collaborative signals
• Connects graph Fourier filters to frequency-domain filters