IMFuse: Instance-Aware Multi-Layer Fusion for LLM-Enhanced Sequential Recommendation

2026-07-29

AI Summary

This work addresses the limitation of existing large language model–based sequential recommendation methods that rely solely on the final-layer representations, thereby overlooking the complementary semantics embedded in intermediate layers and the heterogeneity of items across layer-wise evolution. To this end, the authors propose an instance-aware multi-layer fusion strategy that dynamically integrates multi-layer semantics to generate personalized item representations by learning global dimension-level layer preferences and incorporating an instance-aware expert modulation mechanism. This approach is the first to introduce instance-awareness into multi-layer fusion, effectively mitigating dimensional collapse while balancing global semantic contributions and item-level heterogeneity. Extensive experiments demonstrate that the model achieves an average relative improvement of 6.72% over state-of-the-art methods across four real-world datasets, with manageable parameter and computational overhead.

Abstract

Recent advancements in Large Language Models (LLMs) have significantly enhanced sequential recommendation by encoding rich item textual information into semantic representations. However, existing methods typically rely on the final-layer hidden states of LLMs, overlooking potentially useful semantic signals encoded in other layers. Through empirical analysis, we reveal the limitations of this practice: final-layer representations often suffer from dimensional collapse, whereas intermediate layers preserve complementary, coarse-to-fine semantic knowledge. Furthermore, we observe that different items exhibit heterogeneous layer-wise representation evolution, making a uniform layer selection sub-optimal. To bridge this gap, we propose IMFuse, an instance-aware multi-layer fusion strategy designed for LLM-enhanced recommendation. Instead of relying on a single layer, IMFuse adaptively aggregates multi-layer semantic information by learning global dimension-wise layer preferences to capture general semantic contributions. To address item-level heterogeneity, IMFuse introduces an instance-aware expert modulation mechanism that dynamically adjusts these global preferences, generating personalized, item-specific semantic representations. Extensive experiments across four real-world datasets demonstrate the effectiveness of IMFuse. It consistently outperforms state-of-the-art baselines with an average relative improvement of 6.72%, while introducing limited parameter and computational overhead.

Problem

• sequential recommendation
• Large Language Models
• multi-layer fusion
• dimensional collapse
• representation heterogeneity

Innovation

• Instance-Aware Fusion
• Multi-Layer Representation
• LLM-Enhanced Recommendation
• Dimensional Collapse
• Expert Modulation