基于交互行为相关性的流媒体数据预取策略

Prefetching Strategy of Streaming Data Based on Correlation in Interactive Behavior

用户交互行为是认知流媒体系统的基础和关键技术,将交互行为中的相关性规律建模为隐马尔可夫模型(hidden Markov model,HMM),并由此提出基于隐马尔可夫模型的流媒体数据预取策略.该策略使用Baum-Welch法对HMM的系统参数进行最大似然估计;然后基于HMM模型,利用当前用户的交互行为信息以及过去用户观看状态的后验概率进行贝叶斯推理,对用户当前观看状态的后验概率进行更新;最后根据最大后验概率准则对用户交互过程中的观看状态进行最终判决.使用后验概率,该策略可进一步确定具有最大预取价值的数据块,并实施预取策略以降低视频交互过程中的访问延迟.仿真实验证实了所提策略的有效性.

User interactive behavior is the foundation and key technology of streaming system. Hidden Markov Model （ HMM } was used to model the correlation in interactive behavior, and a prefetching strategy of streaming data based on HMM was proposed. The strategy estimated the HMM system parameters by utilizing a maximum likelihood estimation method called Bantu-Welch algorithm. The posterior probability of the current user browsing state was updated by Bayesian inference, which was based on HMM and de- duced from the posterior probability of the previous browsing state. Finally, the user browsing state was estimated according to the maximum posteriori criterion. Furthermore, the strategy chose the data block which had the greatest caching value and a prefetching algorithm was applied to reduce data access delay. Simulation results show the effectiveness of proposed algorithm.