PulseAugur
EN
LIVE 17:55:37

WHALE architecture unifies recommendation features with Wukong and HSTU

By PulseAugur Editorial · [2 sources] ·

Researchers have introduced WHALE, a novel recommendation architecture designed to efficiently unify non-sequence and sequence features. This model leverages Wukong for high-order non-sequence feature interactions and HSTU for long user-behavior sequence modeling. By integrating these components with an attention-based fusion module, WHALE enables progressive exchange between the two backbones, allowing for detailed retrieval of evidence from user histories. The architecture has been optimized with custom Triton kernels and other system-level co-design techniques to enhance training and inference efficiency, demonstrating consistent offline gains and positive online results in industrial settings. AI

IMPACT This architecture offers a scalable approach to unifying diverse data sources for recommendation systems, potentially improving personalization and user experience.

RANK_REASON The cluster contains an academic paper detailing a new model architecture.

Read on arXiv cs.IR (Information Retrieval) →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

WHALE architecture unifies recommendation features with Wukong and HSTU

COVERAGE [2]

  1. arXiv cs.AI TIER_1 English(EN) · Renqin Cai, Dawei Sun, Yuanjun Yao, Zhiyong Wang, Velvin Fu, Maggie Zhuang, Yu Shi, Zhongnan Fang, Xuan Cao, Jing Qian, Rui Li ·

    WHALE: A Scalable Unified Model for Recommendation with Wukong-HSTU Architecture

    arXiv:2607.17017v1 Announce Type: cross Abstract: As scalability becomes increasingly important in recommendation modeling, recent architectures have advanced the modeling of two broad sources of ranking signals along separate paths: non-sequence features, including user, item, c…

  2. arXiv cs.IR (Information Retrieval) TIER_1 English(EN) · Rui Li ·

    WHALE: A Scalable Unified Model for Recommendation with Wukong-HSTU Architecture

    As scalability becomes increasingly important in recommendation modeling, recent architectures have advanced the modeling of two broad sources of ranking signals along separate paths: non-sequence features, including user, item, context, and cross features; and sequence features …

RELATED ENTITIES

RELATED TOPICS