Analysis of Popularity Pattern of User Generated Contents and Its Application to Content-Aware Networking

In recent years, social multimedia sharing services such as YouTube, which share User Generated Content (UGC) have become much attracted. An efficient control of UGC is one of important roles to achieve, e.g., an optimized placement of advertisements for end users, or content-aware caching control for improving the utilization of network resources. For this reason, it is effective to forecast the future popularity of the content as early as possible, so that we can take a proactive action to highly popular contents. In this paper, we propose a method to classify the popularity of UGCs in real time using K-means clustering, and analyze tendencies led by popularity patterns.We then propose a method to identify UGCs which are expected to be popular in future, by taking both the initial part of popularity patterns and actual counts of content retrieves into consideration. Our experimental results show that the accuracy of identification of popular UGCs can be increased around 10% by considering the initial part of popularity patterns.