Firstly, according to the Hadoop platform the novel data-analysis architecture is designed, then the paper builds the Item-based clustering collaborative filtering algorithm based on Hadoop. And it takes advantage of the MapReduce parallel programming model to improve the traditional collaborative filtering recommendation algorithm, and resolves the problems of poor system performance of traditional personalized recommendation algorithm in high-dimensional sparse matrix operations. Thirdly, this paper elaborates the steps of the parallelization. Lastly the result of the experiment shows that the improved algorithm has obviously better scalability in personalized recommendation for commodity than the traditional Item-based clustering collaborative filtering algorithm.