Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Playback Performance in Video Recommender Systems with an On-Device Gating and Ranking Framework

Yunfei Yang, Zhenghao Qi, Honghuan Wu, Qi Song, Tieyao Zhang, Hao Li, Yimin Tu, Kaiqiao Zhan, Ben Wang

TL;DR

This work addresses playback instability in video recommender systems by moving gating and ranking to the user device. It introduces GRF, a two-stage on-device framework with a gate to flag potential choppy items and a ranking module to substitute them with locally cached videos, coordinating with the cloud RS and a dynamic cache update mechanism. Extensive offline evaluations and online A/B tests on Kwai demonstrate strong gains in both playback stability and engagement, including notable improvements under poor networks and in retention. The results support the viability of on-device, network-free adjustments to enhance user experience in large-scale video platforms and point to broader applicability in edge-enabled RS scenarios.

Abstract

Video recommender systems (RSs) have gained increasing attention in recent years. Existing mainstream RSs focus on optimizing the matching function between users and items. However, we noticed that users frequently encounter playback issues such as slow loading or stuttering while browsing the videos, especially in weak network conditions, which will lead to a subpar browsing experience, and may cause users to leave, even when the video content and recommendations are superior. It is quite a serious issue, yet easily overlooked. To tackle this issue, we propose an on-device Gating and Ranking Framework (GRF) that cooperates with server-side RS. Specifically, we utilize a gate model to identify videos that may have playback issues in real-time, and then we employ a ranking model to select the optimal result from a locally-cached pool to replace the stuttering videos. Our solution has been fully deployed on Kwai, a large-scale short video platform with hundreds of millions of users globally. Moreover, it significantly enhances video playback performance and improves overall user experience and retention rates.

Enhancing Playback Performance in Video Recommender Systems with an On-Device Gating and Ranking Framework

TL;DR

This work addresses playback instability in video recommender systems by moving gating and ranking to the user device. It introduces GRF, a two-stage on-device framework with a gate to flag potential choppy items and a ranking module to substitute them with locally cached videos, coordinating with the cloud RS and a dynamic cache update mechanism. Extensive offline evaluations and online A/B tests on Kwai demonstrate strong gains in both playback stability and engagement, including notable improvements under poor networks and in retention. The results support the viability of on-device, network-free adjustments to enhance user experience in large-scale video platforms and point to broader applicability in edge-enabled RS scenarios.

Abstract

Video recommender systems (RSs) have gained increasing attention in recent years. Existing mainstream RSs focus on optimizing the matching function between users and items. However, we noticed that users frequently encounter playback issues such as slow loading or stuttering while browsing the videos, especially in weak network conditions, which will lead to a subpar browsing experience, and may cause users to leave, even when the video content and recommendations are superior. It is quite a serious issue, yet easily overlooked. To tackle this issue, we propose an on-device Gating and Ranking Framework (GRF) that cooperates with server-side RS. Specifically, we utilize a gate model to identify videos that may have playback issues in real-time, and then we employ a ranking model to select the optimal result from a locally-cached pool to replace the stuttering videos. Our solution has been fully deployed on Kwai, a large-scale short video platform with hundreds of millions of users globally. Moreover, it significantly enhances video playback performance and improves overall user experience and retention rates.
Paper Structure (17 sections, 3 equations, 7 figures, 5 tables)

This paper contains 17 sections, 3 equations, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. RELATED WORK
  3. On-device Recommender Systems
  4. Playback Performance Optimization
  5. SYSTEM
  6. Problem Definition
  7. Stage I : Gating
  8. Stage II : Ranking
  9. Cache update Mechanism
  10. Recommendation in offline mode.
  11. System Implementation
  12. EXPERIMENTS
  13. Offline Results of Gate model
  14. Offline Results of Ranking model
  15. Online A/B Testing
  16. ...and 2 more sections

Figures (7)

  • Figure 1: An example interface of short video platform. (a) Users browse videos by scrolling up and down. (b) and (c) illustrate several playback issues while browsing the videos.
  • Figure 2: The relation between average session watch time and the proportion of choppy videos in a session.
  • Figure 3: The proposed Gating and Ranking framework (GRF) involves two stages: Gating, which screens out choppy video (surround with a red rectangle), and Ranking, which selects the best local cache video for the replacement.
  • Figure 4: The architecture of the on-device Gate model (a) and Ranking model (b).
  • Figure 5: GRF System Implementation.
  • ...and 2 more figures