Assessing the longevity of online videos: A new insight of a video's quality

Recommending valuable videos to viewers is always crucial to an online video website and its related third parties. More particularly, what features and methods to be selected to assess the quality of online videos is still an on-going research topic. Unlike previous work attempted to evaluate a video only by its view count (a.k.a. popularity), this article proposes an additional scoring mechanism to capture a video's long-lasting value (a.k.a. longevity) to assist the judgment of its quality. Generally speaking, a longevous video tends to be watched frequently and therefore is considered to be more valuable. We introduce the concept of latent social impulses and the necessity of using them while measuring a video's longevity. When deriving latent social impulses, we view the video website as a digital signal filter and formulate the task as a least squares problem. The proposed longevity computation is based on the derived social impulses, and we use experiments to directly show that the computed longevity scores are able to capture overlooked information by popularity measure. Unfortunately, the required information to derive social impulses is not always public, which makes a third party unable to directly evaluate all videos' longevities. To solve this problem, we formulate a semi-supervised learning task by using part of videos having known longevity scores to predict the unknown longevity scores, and we propose a Gaussian Random Markov model with Loopy Belief Propagation to solve it. The conducted experiments on YouTube demonstrate that the proposed method can significantly improve the prediction results comparing to baseline models.