Action is All You Need: Dual-Flow Generative Ranking Network for Recommendation
We introduce a Dual-Flow Generative Ranking Network (DFGR) for recommendation scenarios. This architecture utilizes only raw user behavior sequence information together with a small amount of basic information describing the behaviors to address the limitations of Deep Learning Recommendation Models (DLRMs) that rely on extensive manual feature engineering. DFGR employs a dual-flow mechanism to optimize interaction modeling, ensuring efficient training and inference through end-to-end token processing. It duplicates the original user behavior sequence into a real flow and a fake flow based on whether the action information used is authentic and then defines a novel interaction method between the real flow and the fake flow within the QKV module of the self-attention mechanism. This design reduces computational overhead and improves both training efficiency and inference performance compared to Meta's HSTU-based model which can be considered the current state-of-the-art (SOTA) model in generative ranking. Our experiments in open-source and real industrial datasets show that DFGR outperforms DLRM, which can be regarded as an industrial online baseline that uses extensive feature engineering, Meta's HSTU approaches, and common recommendation architectures such as DIN, DCN, DIEN, and DeepFM. We also investigate optimal parameter allocation strategies under computational constraints, establishing DFGR as an efficient and effective next-generation generative ranking paradigm.
View on arXiv@article{guo2025_2505.16752,
title={ Action is All You Need: Dual-Flow Generative Ranking Network for Recommendation },
author={ Hao Guo and Erpeng Xue and Lei Huang and Shichao Wang and Xiaolei Wang and Lei Wang and Jinpeng Wang and Sheng Chen },
journal={arXiv preprint arXiv:2505.16752},
year={ 2025 }
}