TRACER: Balancing Stability-Plasticity-Cognitivity Trilemma for LLM Enhanced Continual Recommendation

2026-08-17

AI Summary

This study addresses the Stability-Plasticity-Cognizability (SPC) trilemma among semantic priors, historical retention, and interest adaptation in LLM-enhanced continual recommendation. We propose TRACER, a novel framework featuring an SPC-responsive adaptive augmentation mechanism. Through a synergistic three-module architecture, TRACER dynamically modulates dimensional weights to prevent single-factor dominance while deeply integrating LLM semantics with user behavioral streams. Extensive experiments on five real-world datasets demonstrate that TRACER effectively reconciles SPC conflicts, achieving up to 14.38% performance improvement over state-of-the-art baselines. These results validate the framework’s capability to facilitate efficient and balanced continual recommendation learning by harmonizing the inherent trade-offs between leveraging large language model knowledge and adapting to evolving user preferences.

Abstract

Continual recommendation aims to capture evolving user interests from streaming data but struggles with sparsity. LLM enhancers mitigate this with semantic knowledge, but naive integration creates a new conflict. We identify this as the Stability-Plasticity-Cognitivity (SPC) Trilemma, where generalized LLM semantic priors (Cognitivity) conflict with retaining personalized historical preferences (Stability) and adapting to individual interest shifts (Plasticity). To address this, we propose Trilemma-Responsive Adaptive Continual Enhancement for Recommendation (TRACER). TRACER synergistically combines three specialized modules, each targeting stability, plasticity, or cognitivity, while preventing any single lemma from dominating. This holistic design enables semantic knowledge to support history retention and adaptation to evolving interests without disrupting continual learning. Across five real-world datasets, TRACER effectively harmonizes the SPC trilemma and outperforms state-of-the-art baselines by up to 14.38%. Our code is available at https://github.com/woo-joo/TRACER_CIKM26.

Problem

• Continual Recommendation
• Stability-Plasticity-Cognitivity Trilemma
• LLM Enhanced Recommendation
• Streaming Data

Innovation

• Stability-Plasticity-Cognitivity Trilemma
• Continual Recommendation
• LLM Enhancement
• Adaptive Continual Learning
• TRACER