A Comparison of Ranking methods for Music Videos Based on an Analysis of Tagging Trend

Abstract

The evaluation of videos in Nico Nico Douga is strongly dependent on some abilities to attract users (e.g., the well-known contributors and the popular contents). Therefore, it is difficult to use the videos which have such abilities and the other videos in the same manner. In contrast, we assume that the user's action that register a certain video into his or her favorites is a Bernoulli trial, and convert the favorite registration rates of each video to the evaluation values which consider the reliability by the number of view counts. Furthermore, to compare the both evaluation methods, we address an extraction problem of categories containing significantly large numbers of highly (or lowly) ranked objects from a dataset of ranked objects with categories. To this end, we newly propose the multi-category order statistics based on an order-preserving mapping. In our experiments, we treat the videos and its tags as objects and categories, respectively.