An Improved Item Based Collaborativ Filtering Algorithm

With the rapid development of Internet technology, which lead to explosive growth of information. While people enjoying rich information resources, they also faces difficulties to the find the information they need from mass of information. In order to help people find the information they wanted efficiently, Recommendation System came into being. Recommendation System is an important information filtering tool, it’s can recommends information to users depending on their historical behavior data, such as rating and comment, but don’t need any explicit demands expression. The key component of Recommendation System is recommend algorithm, and the collaborative filtering algorithm is a successfully-used algorithm for Recommendation System, but the algorithm still faces some challenges, problems such as sparsity, scalability, real-time and so forth. For the item based collaborative filtering algorithm usually performances better, so the paper is based on these algorithm.In the process of item based collaborative filtering, the most critical step is to find item neighbors. But the accuracy of neighbors searching is always limited by the extremely sparse user-item rating matrix, although there are many existing item similarity model can alleviate the problem, there are still some factors haven’t take into accounts. Based on in-depth study to existing similarity calculation model,this paper proposes an integrated item similarity calculation method to alleviate the neighbors searching inaccuracy problem, and then introduces clustering method to modify the traditional collaborative filtering algorithm, which can speed up neighbors finding process without destroying neighbor searching accuracy.Based on the consideration of user’s taste similarity, the difference of average score of different items and the variance of different items, this paper proposes an integrated item similarity calculation method-JAVWeightedModel, For the overwhelming time complexity of JAVWeightedModel, the paper here also introduced K-Center clustering algorithm, which integrated JAVWeightedModel into the clustering process, then we can improve the neighbor finding speed on the promise of neighbor search accuracy.Finally, we validates the correctness and effectiveness of the new method through experiments on different scale MovieLens dataset.