Distillation Matters: Empowering Sequential Recommenders to Match the Performance of Large Language Model

TL;DR

The paper tackles the problem of deploying high-performing LLM-based recommenders with prohibitive latency by distilling their knowledge into lightweight sequential models. It introduces DLLM2Rec, a two-module distillation framework comprising importance-aware ranking distillation and collaborative embedding distillation to transfer reliable, student-friendly knowledge while preserving collaborative signals. Extensive experiments on three real-world datasets show that DLLM2Rec significantly boosts lightweight models (avg. 47.97% over baselines) and can even surpass the LLM teacher in some cases, while maintaining fast inference. The work demonstrates a practical pathway to deploy fast, accurate recommendations at scale without sacrificing the semantic reasoning benefits of LLMs.

Abstract

Owing to their powerful semantic reasoning capabilities, Large Language Models (LLMs) have been effectively utilized as recommenders, achieving impressive performance. However, the high inference latency of LLMs significantly restricts their practical deployment. To address this issue, this work investigates knowledge distillation from cumbersome LLM-based recommendation models to lightweight conventional sequential models. It encounters three challenges: 1) the teacher's knowledge may not always be reliable; 2) the capacity gap between the teacher and student makes it difficult for the student to assimilate the teacher's knowledge; 3) divergence in semantic space poses a challenge to distill the knowledge from embeddings. To tackle these challenges, this work proposes a novel distillation strategy, DLLM2Rec, specifically tailored for knowledge distillation from LLM-based recommendation models to conventional sequential models. DLLM2Rec comprises: 1) Importance-aware ranking distillation, which filters reliable and student-friendly knowledge by weighting instances according to teacher confidence and student-teacher consistency; 2) Collaborative embedding distillation integrates knowledge from teacher embeddings with collaborative signals mined from the data. Extensive experiments demonstrate the effectiveness of the proposed DLLM2Rec, boosting three typical sequential models with an average improvement of 47.97%, even enabling them to surpass LLM-based recommenders in some cases.

Figures (2)

• Figure 1: Illustration the proposed DLLM2Rec that distills the knowledge from the LLM-based recommenders to the conventional recommenders, with leveraging importance-aware ranking distillation and collaborative embedding distillation.
• Figure 2: Sensitivity analysis w.r.t. $\lambda_{d}$, $\gamma_{p}$, $\gamma_{c}$ and $\gamma_{o}$.