Clustering Nico Nico Douga Videos by Using the Distribution of Time-Synchronyzed Comments

Nico Nico Douga (NND) is one of the most famous social media platforms for sharing videos in Japan. There are 17 million videos posted to NND, for which users search by using keyword search and related video recommendation. However, it is difficult for the users to find interesting videos because videos generally are associated with only short texts and a few tags. In this paper, we present a method for clustering videos in NND by using the distribution of time-synchronized comments. Our method regards a video as the set of the users commenting on it and enables the clustering of videos based on users' shared interests. In our experiment, we applied the proposed method to videos posted to NND, and evaluated our method by quantitatively comparing it with existing text-based methods and by qualitatively performing the subjective evaluation of clustering results. In the result of the quantitative evaluation, the proposed method showed a higher score of normalized mutual information than the existing methods when results of text-based methods and category metadata were used as correct results. The experimental results of the qualitative evaluation showed that the proposed method was as good as or better than the existing text-based methods.