Evolving recommender systems with reusable skill modules improves predictions
EvoSkillRec: Skill-Genome Evolution for Recommender Architecture Discovery
Information Retrieval
Summary
Recommender systems help suggest things you might like, but building the best system can be tricky because they need special designs for each kind of task. The authors developed EvoSkillRec, a method that breaks down recommenders into small reusable skill pieces, which can be evolved and combined to create better architectures automatically. It uses both guided changes and new code invention powered by AI to improve the designs step by step. Their tests show this approach works well on various recommendation tasks, balancing prediction quality and computational cost.
What this means in practice
- •For machine learning engineers: Automatically design and optimize recommender model architectures by evolving and reusing validated skill modules for diverse tasks.
- •For software developers in ai: Build AI systems that generate and refine specialized functional components in recommendation software using a guided code evolution approach.
Authors
Xiaopeng Li, Kuo Cai, Bo Chen, Wenlin Zhang, Mengyang Ma, Yingyi Zhang, Zichuan Fu, Yu Yang, Qidong Liu, Yiyu Wang, Ruiming Tang, Wenwu Ou, Jiang Wu, Zhanbo Xu, Xiangyu Zhao
Abstract
Modern recommender systems advance not only by scaling data and parameters, but also by encoding task-specific inductive biases through architecture, including sparse feature interactions for click-through rate (CTR) prediction, temporal attention for sequential recommendation, and expert routing for multi-task learning. However, these biases are typically human expert designed or searched within predefined operator spaces. Although Recent LLM-driven code evolution expands this space, unconstrained edits often produce invalid or ineffective architectures, underuse established architecture design knowledge, and fail to preserve successful innovations for reuse. We introduce EvoSkillRec, a promotion-and-reuse framework for cumulative recommender architecture evolution. It first decomposes recommenders into atomic executable skills and represents architectures as typed skill genomes, with each skill equipped with input--output types, semantic annotations, and implementation code. We then evolve models with different tasks through two coupled spaces: a constrained skill--space that mutates, recombines, specializes, and reuses validated skills, and an open-ended code--space in which LLM planners and synthesizers invent new skill modules using prior evolution traces and accumulated experience. An autoresearch controller evaluates candidates, diagnoses failures, retrieves relevant skills, promotes validated innovations into the skill library, and adaptively allocates the proposal budget between the two spaces. Extensive experiments on CTR prediction, multi-task learning, and multi-domain learning, including resource-constrained co-optimization of predictive quality and model FLOPs utilization in generative ranking models, consistently demonstrate the effectiveness of our proposed EvoSkillRec.