Research Of Collaborative Filter Recommendation Algorithm On Mahout Platform

With the rapid development of the Internet,there is a wide variety of goods on the shopping site so that it is difficult for users to find goods they are interested in.In this context,the recommender system arises at the historic moment.The Recommender system is recommended to users,which can help users to quickly find their valuable information.It has a wide range of application prospects,and is highly concerned by the academic and business circles.Collaborative filtering recommendation algorithm is the most widely used and the most successful recommendation algorithm,but with the continuous expansion of e-commerce sites,traditional collaborative filtering algorithm also faces severe challenges.This paper regards collaborative filtering recommendation algorithm as the research object and finds a way to solve data sparsity and inaccuracy of the similarity calculation.So it makes some research and exploration of collaborative filtering recommendation algorithm from the following aspects:1.Describes overall development of recommender system and recommender algorithm in detail,summarizes the existing recommendation algorithm's and points out respective characteristics as well as compares their merits and shortcomings;this paper analyses in details collaborative filtering algorithm and summary of problems and solutions,which lays a theoretical foundation for the next step research.2.This paper analysis the deficiencies of collaborative filtering recommendation algorithm.To solve the recommendation low-accuracy caused by sparse data,it proposes a new combination of collaborative filtering recommendation algorithm.The algorithm adopts cascade hybrid strategy and matrix filling technology to combine user-based collaborative filtering and item-based collaborative filtering.First of all,it holds that the selection of the nearest neighbor is not reasonable when the traditional collaborative filtering algorithm faces to sparse data,but that is reasonable by using User-based CF to predict user ratings filling the original score matrix leading to reducing data sparse and getting more accuracy of choosing the nearest neighbors;Second,the traditional collaborative filtering algorithm is only considering the similarity between users or items,and similarity calculation is not accurate.However,the new combination of collaborative filtering recommendation algorithm is based on the filled score matrix and adopts Item-based CF for the users to generate recommendation.It fully integrates the advantages of both traditional collaborative filtering algorithm.Through the Mahout platform simulation experiments show that the proposed combination recommendation algorithm,the precision is superior to the traditional collaborative filtering recommendation algorithm.From above research work,in a certain extent,it solves data sparsity problem faced by collaborative filtering recommendation algorithms and makes a great improvement on theory and application research.