Analyzing the highlight prediction of live streaming content

Abstract

With the rise of online live streaming, the highlight prediction has been one of the important problems for both streaming content creators and consumers. This study suggests the way to utilize clip, user- edited moment, to predict highlights of the live streaming videos. Since clips with high aggregated view counts can be seen as a user-defined highlight of the corresponding video, we aim to understand the characteristics of highlight based on popular clips. From the understanding of clip characteristics, we propose the clip features about audience reaction to capture the highlight of live streaming video. We conduct two experiments to validate the effectiveness of the proposed features. The first experiment is to classify the given moment is whether popular clips or not. The second experiment is to predict the highlight moment from the given stream video. We show that our features can predict the highlight moment better than those from previous works. We expect our suggested clip features can be used as an effective signal in predicting live streaming content.