arXiv subjects

Haolei Pei

Publications and source records attributed to Haolei Pei.

3 recordsLinked to original sources

arXiv2026

OneTrans-V2: Unifying Retrieval, Pre-rank, and Fine-rank with One Transformer in Industrial Recommender

Industrial recommendation systems typically operate as a \emph{cascade} of retrieval, pre-rank, and fine-rank, but these stages are usually trained and served as separate models, causing repeated user-sequence encoding, isolated optimization, and duplicated engineering effort. Building on OneTrans' model-level unification, we present OneTrans-V2, one Transformer that unifies the entire cascade. It encodes the user behavior sequence once as a shared context while preserving stage-specific candidate features and computation. Joint training lets the three stages reinforce one another and enables in-model knowledge distillation from fine-rank to pre-rank. We scale the shared backbone with sparse mixture-of-experts (MoE), which increases capacity with bounded activated computation, and stabilize scaling with $μ$P-style parameterization. To consolidate objective-specific retrieval channels, we introduce Decision-Conditioned Generative Retrieval (DCGR). DCGR predicts a decision prefix describing the upcoming interaction and generates items conditioned on it, allowing business objectives to steer a single generative process. Finally, Sequence-Native Training (SNT) organizes training around each user's lifelong behavior sequence and amortizes its encoding across exposures. Deployed across all three stages of a large-scale industrial recommendation system, OneTrans-V2 improves gross merchandise value (GMV) by 9.74\% and, with a co-designed serving stack, delivers $3.2\times$ the throughput of the cascade it replaces under the same hardware budget.

cs.IR↗

arXiv2026

OneTrans: Unified Feature Interaction and Sequence Modeling with One Transformer in Industrial Recommender

In recommendation systems, scaling up feature-interaction modules (e.g., Wukong, RankMixer) or user-behavior sequence modules (e.g., LONGER) has achieved notable success. However, these efforts typically proceed on separate tracks, which not only hinders bidirectional information exchange but also prevents unified optimization and scaling. In this paper, we propose OneTrans, a unified Transformer backbone that simultaneously performs user-behavior sequence modeling and feature interaction. OneTrans employs a unified tokenizer to convert both sequential and non-sequential attributes into a single token sequence. The stacked OneTrans blocks share parameters across similar sequential tokens while assigning token-specific parameters to non-sequential tokens. Through causal attention and cross-request KV caching, OneTrans enables precomputation and caching of intermediate representations, significantly reducing computational costs during both training and inference. Experimental results on industrial-scale datasets demonstrate that OneTrans scales efficiently with increasing parameters, consistently outperforms strong baselines, and yields a 5.68% lift in per-user GMV in online A/B tests.

cs.IR↗

arXiv2024

RimiRec: Modeling Refined Multi-interest in Hierarchical Structure for Recommendation

Industrial recommender systems usually consist of the retrieval stage and the ranking stage, to handle the billion-scale of users and items. The retrieval stage retrieves candidate items relevant to user interests for recommendations and has attracted much attention. Frequently, a user shows refined multi-interests in a hierarchical structure. For example, a user likes Conan and Kuroba Kaito, which are the roles in hierarchical structure "Animation, Japanese Animation, Detective Conan". However, most existing methods ignore this hierarchical nature, and simply average the fine-grained interest information. Therefore, we propose a novel two-stage approach to explicitly modeling refined multi-interest in a hierarchical structure for recommendation. In the first hierarchical multi-interest mining stage, the hierarchical clustering and transformer-based model adaptively generate circles or sub-circles that users are interested in. In the second stage, the partition of retrieval space allows the EBR models to deal only with items within each circle and accurately capture users' refined interests. Experimental results show that the proposed approach achieves state-of-the-art performance. Our framework has also been deployed at Lofter.

cs.IR↗