Predicting the Geographic Distribution of Videos Views from Tags

Prédire la distribution géographique de vidéos à partir de leurs tags

Adrien Luxey (1)

1 ASAP - As Scalable As Possible: foundations of large scale dynamic distributed systems

User Generated Content (UGC) plays a major role in today's Web. People create and share a lot of multimedia content (like photos or videos), leading to heterogeneous and unpredictable distributions of the consumption of such content. To this end, ensuring an acceptable Quality of Service (QoS) to the end-user has become a major challenge for UGC sharing services. Content Delivery Networks (CDNs) provide replication servers that help bring the content closer to its public, through caching mechanisms and proactive placement of the content (that exploit a priori knowledge on the content to predict its viewing distribution). Some approaches study the sharing patterns of a video in social networks, or the trendiness of a video's topic by looking at mainstream media. We would like to propose a proactive placement technique that would only rely on data available to the UGC service, and particularly on the tags associated with content. During the forthcoming internship, we will study the prediction power of Youtube videos' tags on the geographic distribution of its views, how tags can be used for proactive placement in a CDN, and which machine learning techniques apply for this task. This paper will thus present the architectures of UGC systems, the existing placement mechanisms of content on a CDN, and self-contained solutions for proactive placement; including the prediction power of tags for the geographic distribution of the views of a Youtube video.

Mots clés: Inference and natural language processing; Tagset

Domaines: Réseaux sociaux et d'information [cs.SI]

Adrien Luxey. Predicting the Geographic Distribution of Videos Views from Tags. [Internship report] Ecole Normale Supérieure de Rennes; Université Rennes 1. 2016. ⟨hal-01651188⟩