Topic Discovery And Visualization Of Web Video Using Star-structured K-partite Graph

With the development of technology, a large amount of web video websites arise on the internet, such as YouTube, Tudou, Youku, CNTV and so on. Nowadays, people can share and browse videos easily, which makes the number of video increase at an accelerated speed. It is announced that users of YouTube upload about 65,000 new videos and view more than 100 million videos each day. Due to these videos, users can get much useful information while problems happen on the other hand. There are so many videos that it is hard for users to discover the hot topics and events hidden in these videos.Considering all above situations, this paper proposes an automatic topic discovery technology using star-structured K-partite graph, which aims to help users discover the topics hidden in numerous web videos. Firstly, analyze the raw videos and extract all kinds of features, such as visual, textual and audio. Secondly, construct a star-structured K-partite graph to represent the relationships between videos and each kind of feature. In the end, a co-clustering algorithm is employed to discover the topics hidden in the videos.In addition, a system framework is also proposed based on topic discovery technology in this paper. The framework includes topic discovery, topic ranking and topic visualization, which can give users obtain a global overview of all topics at the first glance and make decisions on whether to take local browsing of interested topics through user interactions.