Event Detection on Online Videos Using Crowdsourced Time-Sync Comment

In recent years, more and more people are like to watch videos online because of its convenience and social features. Due to the limit of entertainment time, there is a new requirement that people prefer to watch some hot video segments rather than an entire video. However, it is a quite time-consuming work to extract the highlight segments in videos manually because the number of videos uploaded to the internet is huge. In this paper, we propose a model of event detection on videos using Time-Sync comments provided by online users. In the model, three features of Time-Sync comments are extracted firstly. Then, user behavior relevance in time series are analyzed to find the video shots that people are interested in most. Metric and its optimization to score video shots for event detection are introduced lastly. Experiments on several movies shows that the events detected by our method coincide with the highlights in the movies. Experiments on movies show that the events detected by our method coincide with the highlights in the movies.