Research Of Recommendation System Based On Data Stream Mining

With the development of Internet technology, more and more people choose to shop online, see the news online and watch movies online. But people do not know what to choose among various options, thus the phenomenon of the so-called "information overload" appears.So it is necessary to filter the information to improve people’s feelings when they are surfing online and help people to find the information what they really want. Thus recommendation system arises at the historic moment.At the same time, because of more and more data appearing in the form of flow, and the utilization of the data stream is not very high, this paper presents an idea to combine the analysis of data stream and recommendation system to get more personalized recommendation results.This paper is aimed to analyse the click stream data. On the basis of this, I research the recommendation method from two aspects. Firstly, I research the algorithm of frequent itemsets mining based on the parallel computing framework of MapReduce, and propose a new parallel algorithm for mining frequent itemsets based on binary-tree(hereinafter referred to as FIMB algorithm). The algorithm can complete the mining work by using a process of Map/Reduce. It fully uses the parallelism of the cluster without iterating calculation, and it can provide the result of related recommendations according to the result of mining. Secondly, this paper puts forward a recommendation system model based on trust in social networks. The main idea of this model is that the agents get needed information in the social network and filter the information through the trust relationship between the agents. At the same time, I find that the density of network, preference heterogeneity among agents and knowledge sparseness are the crucial factors affecting system performance. Finally, this paper combines the analysis of click stream data and two recommendation models to propose a recommendation system framework model based on clickstream, and the paper gives the system framework in the form of the flow chart. In the last of this paper, there are several criteria used to evaluate the effect of the recommendation system.In this paper, the efficiency of FIMB algorithm is proved be higher than CD and DD algorithms by the way of contrasting in the experiment. And then I use the mathematical formula to approximately analyze the effect of recommendations based on the trust relationship in social networks. In conclusion, the research in this paper has certain research value and innovation.