TGR-EvoDAG improves generative recommendation with multi-agent code evolution
TGR-EvoDAG: A DAG-based Multi-Agent Code Evolution Framework for Generative Recommendation on Tencent
Artificial Intelligence
Summary
It can be hard to improve recommendation systems that generate suggestions by evolving the underlying code, especially when the process is complex and costly. The authors developed TGR-EvoDAG, a system that treats code improvements as a graph of ideas instead of a simple line, allowing better tracking and combining of different improvements. It also plans computing resources better and carefully evaluates changes to avoid dropping good ideas too soon. Tests on public data and Tencent’s platform showed notable improvements in recommendation quality and user engagement.
What this means in practice
- •For recommendation system engineers: Use multi-agent DAG-based code evolution to autonomously improve generative recommendation pipelines with cost-aware resource management.
- •For large-scale media platform operators: Deploy evolved generative recommendation algorithms that improve click-through rates and offline accuracy in commercial environments.$Commercial implications: Provides a commercially viable generative recommendation pipeline enhancement boosting user engagement on media platforms.
Authors
Junlin He, Junzhang Zhu, Yudong Li, Yihong Tang, Xuefeng Jiang, Wei Ma, Chengxiang Zhuo, Zang Li
Abstract
Large Language Model (LLM) agents have enabled automated code evolution for algorithm discovery. However, current recursive self-improvement (RSI) research for sequential recommendation faces a dilemma: general-purpose frameworks lack domain-specific feedback, while recommender-specific frameworks rely on linear evolution. Crucially, naively integrating them fails when applied to Generative Recommendation (GR). Unlike end-to-end models, GR relies on a module-cascaded pipeline (tokenization, pre-training, and alignment). In this paradigm, existing evolution methods suffer from: i) cost blindness, squandering budgets by ignoring stage-asymmetric retraining overheads; and ii) short-sighted evaluation, discarding promising upstream innovations that temporarily regress metrics before downstream adaptation. To bridge these gaps, we propose TGR-EvoDAG, a multi-agent code evolution framework tailored for GR. TGR-EvoDAG models code evolution as a Directed Acyclic Graph (DAG) to maintain multiple evolution branches and merge complementary innovations, unifying advanced search with rich GR diagnostic feedback. We further introduce stage-aware cost planning to optimize GPU-hour budgets, and active downstream co-adaptation to navigate temporary regressions and unlock cross-stage gains. On public Amazon datasets, TGR-EvoDAG enables the first autonomous evolution of a complete GR pipeline, improving offline metrics by up to 37.4% and Agent4Rec simulated user satisfaction by 19.7% over the strongest baseline. Furthermore, in industrial deployment on Tencent's large-scale commercial media platform, the evolved GR achieves a 12.16% relative improvement in offline Hit@1 and a statistically significant 0.40% increase in online click-through rate (CTR).