Research On Streaming Pre-fetching Technology Oriented To Social Network

With the advent of Web2.0 era, as the representative to the YouTube, video sites based on user-generated content (UGCs) have been highly sought after users on the Internet. But related studies have shown as the impact of restricted network bandwidth resources and user scale, delay existing in Internet video playback stream generated higher proportion of the length of the video, which directly affect the quality of the user experience.To reduce video playback delay, in recent years, the industry put forward a variety of streaming media system architecture optimized solutions from different angles, such as improving server performance of hardware and software, increasing network bandwidth, deploying proxy and build content distribution network(CDN). Although these programs can ease video playback delay, but also increase the deployment costs and lack of scalability. To this end, academia and industry carry out the research of prefix prefetching techniques from the perspective of the video software. However, due to the video service based on user-generated content has many characteristics different from traditional video services on user behavior and video content, such as the larger number of videos, the shorter video length, less content data, faster generate content speed, the popularity of the uneven distribution. Thus traditional video prefetch strategy is not well applied to user-generated content video services system.To reduce user video response latency and to improve the quality of the viewing experience for users, we considers the characteristics of videos with different popularity, mining the social relationships between users and related relationship between videos, integrating both video popularity and information of user historical in user-generated content video services site YouTube, and carrys out research to the prefetching strategies based on the popularity of videos and social networking information. We also use real data sets to analysis the performance of algorithm proposed in the paper. The main contributions of this paper are summarized as follows:1) We use YouTube API to get a YouTube dataset composed of user information, video information and user behavior information. Through analysis of the data set, we find the popularity distribution of YouTube videos approximately follows the Zipf s law with the exponent cut, which indicates the popularity of YouTube videos showing "the rich get richer" phenomenon and the number of unpopular videos is significantly less than the predicted results of Zipf law. Therefore, YouTube videos can be divided into two types of popular video and long-tail video.2) As the popularity of popular video is very high, we proposed a popular video prefetching algorithms combination of popularity and user interests; As the popularity of long-tail video is not high, we anslyse the topology feature of social networks in YouTube, and propose a graph model based on social network and users’ explict feedback. Based on the graph model we proposed, we design a algorithm to measure the similarity between user node and video node, and prefetch long-tail videos with high similarity to the user.3) Due to the characteristics of diversity and difference of users’ watching behaviors, we proposed a personalized mixing prefetch model to fuse popular videos and long-tail videos based on the historical information of users’ feedback.