Design And Implementation Of Dynamic User Behavior Prediction System Based On Movie Score Data

With the rapid development of the Internet and the continuous expansion of the user scale,a variety of applications on the Internet,including film application platform,e-commerce sites,and forums have been rapid development.The increasingly perfect user experience continues to attract people to share and score,which results in massive scoring data that provides valuable data resources for analyzing user behavior and finding user preferences.Taking into account the dynamic factors of user scoring behavior and the challenge of capturing dynamic user behavior,this paper uses the Hadoop Mapreduce framework to design and develop a kind of system which can dig the dynamic user scoring behavior from the massive film user scoring data with timing sequence based on the Bayesian network.The system consists of four major modules,including The storage of massive movie scoring data and Bayesian network,Patching the missing value of hidden variables,The construction of user scoring behavior prediction model in time slice,The construction of dynamic user scoring behavior prediction model between adjacent time slices.Brief analysis is as follows:(1)The storage of massive movie scoring data and Bayesian Network.We use MapReduce parallel programming model to read the massive movie scoring data,and Bayesian network with directed acyclic graph structure,conditional probability table corresponding to each node,then deposit them in HBase by taking the score attribute as the row identifier and the corresponding probability as the column family.(2)Patching the missing value of hidden variables based on MapReduce.We put the expectation maximization algorithm parallelized,and will get a complete set of movie scoring data with hidden variables by computing.(3)The construction of user scoring behavior prediction model in time slice based on MapReduce.We combine the Climbing Method,Bayesian Information Criterion with MapReduce to complete the construction of user scoring behavior prediction model in time slice efficiently.(4)The construction of dynamic user scoring behavior prediction model between adjacent time slices based on MapReduce.Based on the results of step(3),we add the mutual information dependence measure,and finally get the prediction model of the user behavior between adjacent time slices,then predict the behavior of users in the next time slice according to the constructed model.Experiments based on the MovieLens data set show that the prediction system based on dynamic behavior is effective and efficient in predicting the behavior of users.