Recommendation System Practice Based On Mahout And Algorithm Improvement

Nowadays society has entered the era of information explosion, facing a large amount of information and data, whether the producers or the consumers of information are subject to a great deal of information challenges. On the one hand, information producers hope their own information be pushed to the people who may be interested in; on the other hand, the information consumers want to be able to find out what they are really interested in from the vast information. In this situation, the recommended system came into being. In order to model the users' interests, users' historical behavior information will be analyzed by recommendation system. By doing so, recommendation system will be able to predict potential items that the users may be interested in, complete the personalized recommendation.In the field of personalized recommendation system, how could make the performance of the recommendation system achieve optimal is most concerned about.Recommendation algorithm plays a vital role in the quality, so the study on recommendation algorithm has become a focus of attention.The paper is expounded from the following aspects :At first, in order to understand the recommendation system more specific and profound,the concept of recommendation system,building blocks, evaluation criteria, evaluation methods as well as typical application scenarios are introduced. Then several typical recommendation algorithms, including user-based collaborative filtering algorithm,Slope One algorithm, SVD algorithm and LFM are analyzed systemically,the ideas of these algorithms ?algorithm steps as well as the advantages and disadvantages of these algorithms are summarized respectively.Then, the knowledge about the recommended parts of the Mahout is mainly discussed,on this basis, a simple recommendation system based Mahout is built.The recommended platform is based on a single memory, which can process the data below 1M.Subsequently,the platform is used to evaluate the performance of the proposed algorithms, the main evaluation criteria are MAE, Recall and Precision.In the process of simulation, it is found that for different algorithms, when the selected parameters and data sets are not the same,the evaluation results will also be changed.Finally, according to the characteristics of video recommendation system, The shortcomings of the traditional recommendation algorithm in the field of video recommendation are pointed out.Next,Item-based-CF and ALS-WR are introduced in detail.The shortcomings and advantages of these two algorithms are analyzed,too.And on this basis, a hybrid recommendation algorithm weighted combination of these two algorithms is proposed, at last how to determine weighting coefficients is discussed.The simulation results show that the proposed algorithm can effectively solve the problem of poor recommendation accuracy caused by data sparse, and provides the recommendation can meet the user's personalized needs.In short,it will be useful to the research on video recommendation algorithm.