The Research Of Distributed Collaborative Filtering Recommendation System For Cold-Start

Collaborative filtering technology can predict potential interest value of the target object for users based on historical evaluation of information data by users, and then be used as a basis to provide the content of interest for target users. The traditional collaborative filtering technology is facing data sparsity, cold start, scalability and other problems with the continuing growth of forecast data size, and these problems have resulted in the decrease of effectiveness and efficiency of the algorithm, then even make the user experience a serious imbalance.Aiming at cold start problems of data sparse in the algorithm, this paper does further research which includes three aspects:Firstly, collaborative filtering recommendation algorithm of the tag technology and two-way clusters are proposed. The algorithm, which uses the relationship of the tags after clustering and tags-items, first predicts the content ratings. Then, it divides the users and the contents into one more categories according to probability, and fills each category with no rating contents. At last, the algorithm recommends target object for users based on content similarity calculation method.Secondly, clustering algorithm and recommendation algorithm which are two key contents of the paper’s algorithm are designed and implemented in a distributed framework. In the algorithm, both the above two algorithms are very consuming resources. By applying Map Reduce framework based on Hadoop platform, this paper, which programs in a distributed computing environment for the united clustering and recommendation algorithm, can run an efficient and accurate.Finally, the experiments prove that the algorithm has a greater advantage in solving the cold start problem compared with several existing algorithms. The algorithm which designed and implemented by distributed framework can handle massive target effectively and shorten operation hours of the recommended system significantly.