Home > Research > Publications & Outputs > Forecasting audience increase on YouTube
View graph of relations

Forecasting audience increase on YouTube

Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSNConference contribution/Paperpeer-review

Published
Publication date30/05/2011
Host publicationProceedings of the International Workshop on User Profile Data on the Social Semantic Web co-located with 8th Extended Semantic Web Conference May 30, 2011, Heraklion, Crete, Greece
EditorsFabian Abel, Qi Gao, Eelco Herder, Geert-Jan Houben, Daniel Olmedilla, Alexandre Passant
Number of pages15
<mark>Original language</mark>English
EventUser Profile Data on the Social Semantic Web (UWeb2011) Workshop, Extended Semantic Web Conference 2011 - , Greece
Duration: 30/05/2011 → …

Workshop

WorkshopUser Profile Data on the Social Semantic Web (UWeb2011) Workshop, Extended Semantic Web Conference 2011
Country/TerritoryGreece
Period30/05/11 → …

Workshop

WorkshopUser Profile Data on the Social Semantic Web (UWeb2011) Workshop, Extended Semantic Web Conference 2011
Country/TerritoryGreece
Period30/05/11 → …

Abstract

User profiles constructed on Social Web platforms are often motivated by the need to maximise user reputation within a community.
Subscriber, or follower, counts are an indicator of the influence and standing
that the user has, where greater values indicate a greater perception or regard for what the user has to say or share. However, at present there lacks an understanding of the factors that lead to an increase in such audience levels, and how a user’s behaviour can affect their reputation. In this paper we attempt to fill this gap, by examining data collected from YouTube over regular time intervals. We explore the correlation between the subscriber counts and several behaviour features - extracted from both the user’s profile and the content they have shared. Through the use of a Multiple Linear Regression model we are able to forecast the audience levels that users will yield based on observed behaviour. Combining such a model with an exhaustive feature selection process, we yield
statistically significant performance over a baseline model containing all
features.