Research And Implementation Of Content Acceleration System Based On User Behavior

With the rapid development of the Internet, the demand of users for high definition(HD) video has been increasing. It has been a hot trend on application to provide HD video for large scale users. Currently the network access bandwidth or server capacity may become a bottleneck, which results in long-time video stream buffering time, intermittent playback and poor user experience.Taking advantage of the text of web pages that user browsed and the film rating of the user, a content acceleration system is realized based on user behavior in this thesis. The system combines user-based collaborative filtering recommendation method and content-based recommendation method to recommend videos for users. Then download and manage the recommended videos. When a user wants to watch the video, he can watch from the local directly, avoiding waiting too long for buffering. At first, the thesis designs the entire structure of the system, the whole design is divided into two parts: recommendation subsystem based on user behavior and content acceleration subsystem. Then design and implement the two parts. The recommendation subsystem based on user behavior focus on the methods of hybrid recommendation and users' similarity calculation. Content acceleration subsystem is mainly file management, file download, file mapping and redirect research. Finally, a lot of experiments have been conducted to test the performance of the system in terms of different indexes and the results are analyzed as well. The main contributions of this thesis are as follows:(1) When calculating the similarity between users, we design a user rating sparse actor in the traditional method, which can effectively alleviate the phenomenonof sparse user ratings, which leads to the lack of accurate calculation of users' imilarity.(2) Use hybrid recommendation algorithm which combined with the user-based ollaborative filtering and content-based recommendation method to recommend ideo for the user, improved the accuracy of the recommendation.(3) Content prefetching acceleration. Recommended subsystem calculates the ecommended video list and passes it to content acceleration subsystem. Content cceleration subsystem downloads the recommended videos from the network by prefetching algorithm and stores them in local-storage area. Users access the internet through the proxy server. When users want to watch movies on the local- storage area, we will show the film to the user, so as to achieve acceleration.Finally, the crawler collects a large number of users' diary, users' film scores and films' profile information from douban to simulate user behavior data. Experiment with these data, create test cases for different parameters. calculate the precision, the recall, F1 value through a large number of experiments. Carrying out detailed analysis and evaluation of the implemented system. Using user-based collaborative filtering method as baseline. Test results show that the recommended method in this thesis is better than the common user-based collaborative filtering method in terms of precision, recall and F1 value. They also show content acceleration subsystem is running fluent. So the entire content acceleration system based on user behavior is running fluent and has good user experience.