🤖 AI Summary
In short-video recommendation, user viewing duration exhibits high uncertainty; conventional point-wise (mean) prediction fails to capture behavioral diversity, limiting recommendation effectiveness. This paper proposes a Conditional Quantile Estimation (CQE) framework—the first to systematically integrate deep quantile regression into short-video viewing duration modeling—enabling end-to-end learning of the full conditional duration distribution for user-video pairs. The method comprises joint multi-quantile modeling, an adaptive fusion strategy, and differentiable optimization, thereby departing from the mean-prediction paradigm. Online A/B tests on the Kuaikan mobile application demonstrate statistically significant improvements: +3.2% in active days, +2.8% in active users, +4.1% in average viewing duration per user, and +5.0% in total video plays. These results empirically validate that distributional prediction yields tangible gains in recommendation quality.
📝 Abstract
Accurately predicting watch time is crucial for optimizing recommendations and user experience in short video platforms. However, existing methods that estimate a single average watch time often fail to capture the inherent uncertainty and diversity in user engagement patterns. In this paper, we propose the Conditional Quantile Estimation (CQE) framework to model the entire conditional distribution of watch time. Using quantile regression, CQE characterizes the complex watch-time distribution for each user-video pair, providing a flexible and comprehensive approach to understanding user behavior. We further design multiple strategies to combine the quantile estimates, adapting to different recommendation scenarios and user preferences. Extensive offline experiments and online A/B tests demonstrate the superiority of CQE in watch time prediction and user engagement modeling. In particular, the online deployment of CQE in KuaiShow has led to significant improvements in key evaluation metrics, including active days, active users, engagement duration, and video view counts. These results highlight the practical impact of our proposed approach in enhancing the user experience and overall performance of the short video recommendation system. The code will be released after publication.