From views to tags distribution in Youtube

Stéphane Delbruel, François Taı̈ani2014 · DOI

Abstract

Better understanding the geographic distribution of user generated content (UGC) can help design better distributed implementations of UGC systems. Although tags play a critical role in many social media, their role in the geographical distribution of UGC content has been little investigated. In this paper, we analyze how the tags attached to a video relate to the geographic distribution of a video's views in a YouTube dataset, and suggest how such knowledge might be used to improve current UGC services.

Introduction

Streaming video accounts for a massive portion of global internet traffic, often reaching up to 60% of an ISP's load during peak hours. For platforms like YouTube, the challenge lies in managing the "long tail" of content—millions of niche videos relevant only to specific audiences. Efficiently caching this content requires predicting where a video will be popular before it goes viral. This study explores whether uploader-provided tags can serve as a proxy for predicting future geographic distribution.

Methodology: Reconstructing Geographic Views

The researchers analyzed a dataset of approximately 691,349 YouTube videos representing 173 billion views. Since YouTube provides a normalized "popularity vector" (0–61) rather than raw view counts, the authors developed a reconstruction method using Alexa Internet data to approximate global YouTube traffic distribution.

Global distribution of YouTube traffic estimated from Alexa data.

By combining popularity scores, estimated national traffic, and total view counts, they reconstructed an approximation of views per country. They then aggregated these views across videos sharing the same tags to identify geographic clusters.

Results: Tags as Geographic Anchors

The analysis revealed that semantic tags often act as "geographic anchors." For instance, a tag like "favela" is almost exclusively viewed in Brazil, whereas a tag like "pop" mirrors the general global distribution of YouTube users.

Geographic distribution of views for a localized tag, showing high concentration in specific regions.

This finding suggests a "proactive geo-caching" strategy. Instead of waiting for a video to gain traction, Content Delivery Networks (CDNs) could use tags at the moment of upload to decide where to store data. A video tagged with "favela" could be immediately pushed to South American servers, reducing latency and backbone traffic.

Conclusion and Future Work

While the dataset (from 2011) is dated, the core principle remains relevant: geographic locality is a dominant factor in the consumption of "long tail" content. Future work aims to develop automated algorithms that decide server placement based on the linguistic and semantic properties of tags, potentially revolutionizing how UGC platforms manage global traffic.

Summary updated

Abstract

Better understanding the geographic distribution of user generated content (UGC) can help design better distributed implementations of UGC systems. Although tags play a critical role in many social media, their role in the geographical distribution of UGC content has been little investigated. In this poster, we present some early work analyzing how the tags attached to a video relate to the geographic distribution of a video's views in a Youtube dataset, and hint at how such knowledge might be used to improve current UGC services.
BibTeX
@misc{delbruel2014from,
  author = {Stéphane Delbruel and François Taı̈ani},
  title = {{From views to tags distribution in Youtube}},
  year = {2014},
  doi = {10.1145/2678508.2678519},
}
Figure