Abstract: In P2P streaming media system, if users want to get a better viewing quality, the system will choose the video segment with higher popularity. In some cases, the shortest response time is not the best video segment, it also depends on the needs of users. The popularity of the newly released video segment has not formed a stable trend, so there is not enough data, and traditional statistical methods cannot reflect the changes in popularity in time. To solve this problem, this paper proposes a video prediction caching strategy Modified Markov Prediction Model (MMPM) based on Markov modified model. This strategy can be run when there are not many historical access records of users. It obtains the state transition matrix from the number of clicks on the video segment to adapt to the continuous change of user click rate. Simulation experiments show that the realization of dynamic prediction improves the hit rate and response speed, and verifies the effectiveness, accuracy and speed of the algorithm.

Key words: Markov, modified model, popularity, hit rate