The Research On IPTV Video Hits Prediction Algorithm And System Implementation

In the past few years, IPTV has been widely accepted due to the rapid development of broadband network from ADSL (Asymmetric Digital Subscriber Line) to FTTH (Fiber To The Home), the leap of mobile network technology and the diversified usage of Internet. Users have changed their way of using IPTV from simply watching live channels and time-shifting channels to Video On Demand service. Such change of using method leads to a big evolvement in the program broadcasting mode. Programs are now selected fully according to users’ preferences instead of pushed to the users half-mandatorily.Therefore, how to predict the playback times of each program under this new using method becomes an eager concern for different characters in the program making and playing process, such as producers, VOD (Video On Demand) service providers and advertisers. Prompt and accurate video hits prediction can provide strong support for commercial decision.Traditional times series prediction usually needs huge amount of history data to give out the appropriate prediction result, which is difficult to meet the real needs. Hence, we propose two prediction algorithm SSMP (Shifted Shape Match Prediction) and DSLR (Dynamic Switching Long-term Regression) to make the video hits prediction prompt and accurate. SSMP, with the combination of KNN (K-Nearest-Neighbor) framework and its novel shape match technology, can make full use of a small amount of historical data to produce prompt long-term prediction. It has realized the early prediction, which is a big challenge for traditional methods, as well as significantly outperformed the basic KNN (K Nearest Neighbor) regression. On the other hand, DSLR was built on the basis of SSMP to cover its shortage in continuity, as well as keep the promptness of SSMP, and finally give out a complete continuous predicting solution. Through plenty of experiments on real dataset, it has been proved that DSLR can choose appropriate switching point and utilize the historical data to make continuous and accurate prediction in the stable stage of video hits series. It can provide better effect than SSMP and traditional time series prediction when dealing with the stable stage prediction problem. Through the research on SSMP and DSLR, we’ve realized prompt, continuous and accurate prediction for video hits in different time stage.