Estimation of Tags Using Various Data for Online Videos

We propose a method for estimating new tags to enrich folksonomy on video sharing sites using various video information. Folksonomy is a process by which users tag videos to facilitate searches. One such website is Nico Nico Douga, although users of the site cannot post more than 12 tags to a video. Consequently, although some important tags will be posted, others might be missing. We present a method for acquiring some of these missing tags by choosing new tags. The method uses comments, tags, titles, descriptions, and thumbnails to estimate new tags. Using these data, we develop various algorithms, including neural networks, word embeddings, and K-means. Our method indicated high performance in experimental results.