SmartVideoRanking: Video Search by Mining Emotions from Time-Synchronized Comments

摘要

Many people search for and watch videos on video sharing Web sites, where users input a query and rank videos on the basis of metrics such as view count and rating. However, it is not always easy to find the desired video with a conventional search. One approach that enables users to more intuitively search for videos they desire is to index them according to emotions. Previous studies have used several predefined emotion categories, such as "fear" and "funny", for this purpose. However, viewers' emotions tend to be more diverse and specific. In this paper, we dynamically detect emotions in accordance with an input query and implement SmartVideoRanking, which enables users to search for videos on the basis of the detected emotions. We estimate viewer emotions from time-synchronized comments on videos and estimate the usefulness of each emotion by using support vector machine regression. Experimental results show that: (1) Spearman's rank correlation between the estimated usefulness scores and gold standard data was 0.7547; (2) emotions associated with videos vary from one query to another and it is therefore meaningful to detect emotions according to an input query; and (3) rankings based on viewer emotions enable users to browse videos that do not appear at the top of conventional search results. We also conduct a user study and demonstrate SmartVideoRanking's capability to search for videos.