A First Look at User Switching Behaviors Over Multiple Video Content Providers

Huan Yan, Tzu-Heng Lin, Gang Wang, Yong Li, Haitao Zheng, Depeng Jin, Ben Y. Zhao

Introduction

In the rapidly evolving landscape of digital video, understanding user loyalty and platform competition is crucial. Traditional research often focuses on user behavior within a single platform, such as Netflix or YouTube. However, this "siloed" view fails to capture the reality of "multi-homing," where users jump between various apps and websites to satisfy their content needs. This paper presents a large-scale analysis of how users navigate an ecosystem of multiple video content providers (CPs), debunking the myth of the "loyal" user and revealing the complex dynamics of cross-platform switching.

Methodology and Dataset

The researchers analyzed a massive dataset from a major Chinese ISP, covering 9 million users in Shanghai over two months in late 2015. Using Deep Packet Inspection (DPI), they captured 269 million viewing requests across the "Big Six" platforms: Youku (YK), IQiyi (IQI), Sohu (SH), Kankan (KK), LeTV (LE), and Tencent Video (TC). This ISP-level perspective allowed for a holistic view of user movements that individual platforms cannot see.

To quantify these behaviors, the study introduced metrics such as View Entropy (measuring the spread of views across CPs) and Cross-site Switching Frequency (measuring how often consecutive views occur on different platforms).

User Multi-homing and Switching Patterns

The analysis reveals that active viewers are highly "promiscuous." While nearly half of all users in the dataset visited only one site, this behavior changes drastically with engagement levels. Among heavy users (those watching at least 100 videos a month), nearly 50% visited all six platforms, while only 2% remained loyal to a single provider.

Distribution of the number of websites visited per user, showing a significant portion of users multi-homing.

Despite this multi-homing, users do not constantly flip between apps. The switching frequency is relatively low (typically < 0.5), indicating a "bingeing before jumping" pattern. Users tend to complete a session or watch several videos on one site before migrating to another.

Content Dynamics and Platform Specialization

Content type is a primary driver of both views and switching behavior. TV series, in particular, exert a massive influence, accounting for 41.8% of all views despite representing only 8.4% of the total video library.

Comparison of view percentage vs. video library percentage across content categories for six major CPs.

The data also highlights how smaller CPs survive through specialization. While larger platforms like Youku act as a "gravity well" for switchers due to their library depth, others find success in niches. For instance, Tencent Video leveraged social integrations to dominate "News" views, while Sohu performed strongly in the "Cartoon" category.

Key Insights and Results

  • The Gravity of Scale: The probability of a user switching to a specific CP is highly correlated with that CP’s total market share. Youku, the largest provider, was the most frequent destination for users leaving other platforms.
  • Entropy and Exploration: As users visit more CPs, their view entropy increases, suggesting they distribute their attention relatively evenly rather than just visiting secondary sites occasionally.
  • Social Integration: Platforms that successfully integrated with social networks (like TC with WeChat/QQ) saw distinct advantages in specific content categories.

Limitations and Future Work

The study acknowledges several limitations, including the geographic focus on Shanghai and the temporal relevance of 2015 data, which predates the explosion of short-form video platforms like TikTok. Additionally, the increasing use of HTTPS encryption poses challenges for ISP-level DPI in modern contexts. Future research could focus on modeling the specific triggers for switching—such as the conclusion of a series or a drop in stream quality—to help providers better predict and manage user churn.

Summary updated

Abstract

Watching videos from multiple content providers (CP) has become prevalent. For individual CPs, understanding user video consumption patterns among CPs is critical for improving on-site user experience and CP's opportunity of success. In this paper, based on a two-month dataset recording 9 million users' 269 million video viewing requests over 6 most popular video CPs in China, we provide a first look at users' video consumption and switching behaviors across different CPs. We find that rarely do active video watchers stick to one single CP. When users switch from one CP to another, they are more likely to switch to larger CPs with more videos.
BibTeX
@inproceedings{yan2017first,
  author = {Huan Yan and Tzu-Heng Lin and Gang Wang and Yong Li and Haitao Zheng and Depeng Jin and Ben Y. Zhao},
  title = {{A First Look at User Switching Behaviors Over Multiple Video Content Providers}},
  booktitle = {ICWSM},
  year = {2017},
}
Figure