发表机构
AI VK; Lomonosov MSU（AI VK; 莫斯科大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对短视频观看时长分布预测的EGMN模型存在的缺陷，提出HEGM模型并经实验验证可提升推荐相关性能，且生产测试获显著效果。
AI 中文摘要
准确的观看时长（Watch-Time, WT）预测是短视频推荐的重要需求。然而观看时长分布呈现近零膨胀、长尾且多模态的特性。近期的指数-高斯混合网络（Exponential-Gaussian Mixture Network, EGMN）可对完整的条件观看时长分布建模，而非仅输出单点估计，达到了当前最优性能。本文的大规模复现研究显示，EGMN存在方差崩溃、组件冗余及组件失活的问题。我们提出分层指数-高斯混合模型（Hierarchical Exponential-Gaussian Mixture, HEGM），通过分层跳过观看分解、基于KL散度的方差正则化、结构化初始化、移除强制高斯偏移及熵正则化项来解决这些失效模式。在公开数据集及大规模工业数据集上，HEGM提升了排序精度与阈值事件预测性能，同时保持了有竞争力的单点估计精度，还大幅改善了混合模型的稳定性与可解释性。一项为期1.5个月的生产环境A/B测试证实，该模型可带来具有统计显著性的用户参与度提升。本文的代码与模型已公开于指定URL。
英文摘要
Accurate watch-time (WT) prediction is an important requirement for short-video recommendations. Yet WT distributions are near-zero-inflated, long-tailed and multimodal. The recent Exponential-Gaussian Mixture Network (EGMN) models the full conditional WT distribution rather than a single point estimate and achieves state-of-the-art performance. Our large-scale reproduction study reveals that EGMN is vulnerable to variance collapse, component redundancy, and inactive components. We propose a Hierarchical Exponential-Gaussian Mixture (HEGM) model that addresses these failure modes through a hierarchical skip-watch decomposition, KL-based variance regularization, structured initialization, removing the forced Gaussian shift and the entropy regularizer. Across public and large-scale industrial datasets, HEGM improves ranking accuracy and threshold-event prediction, while maintaining competitive point-estimation accuracy and substantially improving mixture stability and interpretability. A 1.5-month production A/B test confirms statistically significant engagement lifts. Our code and models are publicly released at https://github.com/rw404/HEGM.
Comments16 pages, 8 figures, 7 tables, accepted at IEEE International Conference on Data Mining (ICDM 2026)
