Hierarchical Exponential-Gaussian Mixtures for Watch-Time Distribution Prediction
2026-08-24 • Information Retrieval
Information RetrievalMachine Learning
AI summaryⓘ
The authors studied how to better predict how long people watch short videos, which is tricky because watch times can vary a lot and have unusual patterns. They found problems in a previous model called EGMN, like it sometimes failed to use all its parts properly. To fix this, they created a new model called HEGM that improves prediction accuracy and makes the model more stable and easier to understand. Testing on real datasets and in a live video app showed HEGM could increase user engagement. They also shared their code publicly.
watch-time predictionshort-video recommendationmixture modelsExponential-Gaussian Mixture Networkvariance collapseKL divergenceA/B testingmodel interpretabilitypoint estimationmultimodal distribution
Authors
Sofia Gulevskaia, Mikhail Trapeznikov, Aleksandr Poslavsky, Alexander D'yakonov
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, 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.