Research Into Streaming Media Proxy Cache Replacement Mechanism Based On Collaborative Filtering Recommendation

With the increase of video resource and network user scale, the bandwidth ofstreaming media server and transmission network has become the bottleneck of the Videoon Demand system based on C/S architecture. Streaming media proxy cache is the keytechnology of reducing the load of streaming media server and transmission network. Thecache replacement algorithm is the determining factor of proxy server performance, itneeds to learn which media data would be probably requested in the further. Thecollaborative filtering recommendation algorithm can analysis the historical accessbehavior of users and predict which media data would be probably requested in the further.Therefore, we introduce the collaborative filtering recommendation algorithm to designthe cache replacement algorithm on the basis of comprehensive analysis on the currentresearch in the two areas. The material contents are as follows.Firstly, CFCD(Collaborative Filtering Recommendation Algorithm Combined withDissimilarity) algorithm is proposed. Traditional similarity calculation methods, to someextent, ignore the dissimilarity information which is hidden in the items that only rated byone of the two users. The concept of dissimilarity is defined based on the improved L1Norm, and is introduced to the traditional collaborative filtering recommendationalgorithm. Then, a collaborative filtering recommendation algorithm combined withdissimilarity is proposed.Secondly, Cache Replacement Algorithm Based on CFCD is proposed. The conceptof RD(Recommended Degree) is defined based on the analysis of the recommendationresult. On the basic of RD, the recommended degree and the possibility to be requested ofmedia data is obtained and the RD of each media data is calculated. The cachereplacement algorithm CRA_CFCD(Cache Replacement Algorithm Based on CFCD) isbased on the caching value function, which takes the RD of the media data intoconsideration.Finally, implement simulation program to validate the effectiveness of collaborativefiltering recommendation algorithm and cache replacement algorithm is proposed in this dissertation. MAE(Mean Absolute Error) is selected as evaluating indicator to verify theperformance of CFCD. In addition, the cache replacement algorithm is compared withother typical algorithms from cache hit ratio and start-up latency.