Design And Implementation Of Video On Demand System Based On HLS

With the rapid development of mobile internet technology,mobile network bandwidth is gradually increasing,and the number of mobile user who need information-based short video-on-demand is growing too,a small scale application can receive hundreds of millions of video on demand data generated everyday.So there are so many problems to be solved,like how to make these users experience a better video playback experience? What strategy and content should we use when we recommend video to users? How to extract the available information from these massive amounts of data? This paper will focus on the issues offerd above to study.Nowadays short-video is very hot,because user don't have to spend too much time to get information from a short-video.But under the condition of weak wifi or mobile network,video on demand often buffer,it always takes seconds from click video play button to the first frame of the video was pushed on the player.During watching the video,it may appear buffering too.Sometimes system even get the wrong information of this video.You can't find a video you are interested in while you take much time on scanning video feeds,Rule-based recommended methods are more applicable to new users,and this method can ensure the content a certain degree of freshness.But to users who used system a period of time,they have a certain interest in the accumulation and historical data,rule-based method can't fulfil their requirements.And over time these users can feel disappointed the abandon this system.This issue will focus on solving the problems above.By studying the live protocols,and apply it to video-on-demand.And to make it more suitable to our system,we make some approvement to the live protocol and other optimizations,to decrease play failure and secondary buffer rate.On matter of recommended system,we adopt the method of machine learning,we choose features and calculate CTR through model fusion,clean the data,through feature proccessing,with the fusion of GBDT and LR,calculate the score,push videos based on the score.