Web 2.0 ushers in a wealth of opportunities for filmmakers and individuals to produce, share, and rate videos over the Internet. The creation, promotion, and distribution of videos require an effective resource allocation plan in which the sales and rental revenues offset the production expenses. Ideally, videos that are trendy and frequently played by customers ought to generate more profits than unpopular videos. However, the cost-effective production of videos is a tricky problem. How can a producer determine the likely popularity of a video before it hits the Internet? How should businesses effectively target advertisements on Internet video hosting and social networking websites?

Figueiredo, in this paper, investigates the impact of video referrals on popularity permanence. He explores the patterns in the sharing of over 24,000 videos from YouTube. The well-known spectral clustering algorithm [1,2] is used to delineate the patterns of the videos from their distinctiveness in variables such as the caption categories, first dates of access, total views, favorable ratings, comments, and views attributable to each referrer. Four clusters of videos emerge from the clustering analysis. The graph of total views over a given period is used to illuminate the trends of each cluster of videos. The number of videos, the average view counts, the average rate of change in the view counts, and the average peak number of views are computed to characterize each cluster. Results show that only one video cluster was popular over time without any significant peak, while three additional clusters each had one peak of popularity and then gradually declined.

The author also examines video features as predictors of popularity trends, including the content (category, lifespan, upload date), referrer link (first date, total views), and popularity (total views, comments, and so on). A set of videos with known popularity trends is used to train an exceptionally randomized ensemble trees prediction technique. This prediction model is then applied to classify the popularity lifespan of a set of test videos. In general, the accuracy of the predicted lifespan popularity of the videos improves with an increase in the monitoring period by the classifier function.

This remarkable paper offers great insight into identifying factors associated with the likelihood of an individual video’s popularity. The author succinctly reviews the limitations of existing video popularity prediction studies, and raises a compelling question about how many views a video will receive. Unfortunately, this question is difficult to answer from the clustering results of videos. Nevertheless, the information on the popularity trends of videos at alternative social networking sites should help businesses determine how to allocate advertising resources.