Hierarchical Exponential-Gaussian Mixtures for Watch-Time Distribution Prediction
Abstract
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, structured initialization, removal of the forced Gaussian shift and entropy regularizer and optional KL-based variance regularization for over-parameterized mixtures. Across large-scale industrial and public datasets, under a fixed-backbone head-replacement protocol, HEGM improves ranking accuracy and threshold-event prediction over reproducible lightweight baselines, 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.
Related
- Canonical Exponential-Family Mixtures for Latent Class Choice Models: A Unified Closed-Form EM Framework, with the Gaussian–Geometric Case
- A novel Weighted Exponential-Lindley distribution with Bayesian estimation under progressive Type-II censoring and a Bayesian-Calibrated Variational Autoencoder for survival prediction.
- Dimension-free comparison inequalities between f-divergences for Gaussian location mixtures
- Solving the Ho--Nguyen Conjecture for Gaussian Mixtures: Applications to Mixtures of Experts
- Squarefree values of random polynomials are Gaussian mixtures
BibTeX
@misc{gulevskaia2026hierarchical,
title = {{Hierarchical Exponential-Gaussian Mixtures for Watch-Time Distribution Prediction}},
author = {Sofia Gulevskaia and Mikhail Trapeznikov and Aleksandr Poslavsky and Alexander D'yakonov},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2608.23356},
eprint = {2608.23356},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.23356},
}
RIS
TY - UNPB
TI - Hierarchical Exponential-Gaussian Mixtures for Watch-Time Distribution Prediction
AU - Sofia Gulevskaia
AU - Mikhail Trapeznikov
AU - Aleksandr Poslavsky
AU - Alexander D'yakonov
PY - 2026
DA - 2026/10/21
PB - arXiv
DO - 10.48550/arxiv.2608.23356
UR - https://arxiv.org/abs/2608.23356
AB - 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, structured initialization, removal of the forced Gaussian shift and entropy regularizer and optional KL-based variance regularization for over-parameterized mixtures. Across large-scale industrial and public datasets, under a fixed-backbone head-replacement protocol, HEGM improves ranking accuracy and threshold-event prediction over reproducible lightweight baselines, 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.
ER -