Building hybrid recommendation system based on Hadoop framework

Abstract

As in Real world The Recommender system plays a measure role by which we can recommend the perfect item from a huge set of products. As day to day numbers of users and items of a recommended system is growing rapidly, so as a result Single node machine will take very large time for implementing this recommendation system. Generally collaborative filtering algorithm is widely used in recommended system. To improve the performance and accuracy we have used distributed collaborative filtering recommender algorithm with combining parallel K-Mean clustering algorithm and slope one on Hadoop framework. Apache Hadoop is an open source technology where we can process huge numbers of dataset with commodity hardware. In this paper we have used MapReduce framework and comparing the differences time consumption of common seral hybrid recommendation algorithm with parallel hybrid recommendation algorithm by using different clusters. Here we have applied the experiments by using Each Movie data sets to exploit the advantages of parallel algorithm. Also from the experiments we can get how our improved parallel K-Mean algorithm by using two methods such as distance measure method and initial centroids method which are based on MapReduce can achieve higher accuracy as compare to tradition K-Mean algorithm and also our performance can improve.