EGMN was proposed by the Xiaohongshu research team in August 2025, and the relevant research results were published in the paper "Multi-Granularity Distribution Modeling for Video Watch Time Prediction via Exponential-Gaussian Mixture Network", nominated for Best Paper at RecSys 2025.

The Xiaohongshu research team proposed an Exponential-Gaussian Mixture Network (EGMN) model using a neural network architecture. This network consists of two key modules: a hidden representation encoder and a mixture parameter generator. First, the researchers generate a hidden representation shared across all distribution components. Then, the parameters of each distribution component are estimated based on the hidden representation, and a gating network is applied to perform a weighted mixture of multiple distributions. The researchers conducted extensive offline experiments on public datasets and online A/B testing in the industrial short video delivery scenario of the Xiaohongshu App, demonstrating that EGMN exhibits excellent distribution fitting capabilities at both coarse-grained and fine-grained levels.