Web Video Clustering Based on Emotion Category
Vinath Mekthanavanh, Tianrui Li, Jie Hu, Yan Zhu Yang · 2018
Web video clustering is a fundamental task in the field of social media mining. Automatically web video categorization methods enable users to find video corresponding to their interests. However, all the previous studies are only conducted on the default categories given by the website, i.e., 15 categories of YouTube. To date, clustering based on emotion category has not been a factor considered in this area. Therefore, in this paper, we propose a method to cluster YouTube videos into six emotion categories (e.g., angry, disgust, happy, horror, sad, surprise) with expect to improve video search results. The YouTube data is collected. Word embedding is utilized for transforming the video document into vectors which are then used in a clustering task. Clustering ensemble is employed to obtain final results. We compare the performance of this method with a state-of-the-art technology, i.e., Term Frequency-Inverse Document Frequency based on Vector Space Model. The experiment is implemented on a benchmark dataset for web video analysis. The results show that the best performance is achieved by applying clustering ensemble which reflects the feasibility of clustering web videos into suitable emotion categories.