Research And Implementation Of Recommender System Based On Hadoop And Mahout

Rapid development of Internet leads to information explosion. In electronic commerce, information overload causes bad shopping experiences, leading to customers churn. To find points of interest quickly and to make recommendation are main challenges faced by marketers overall. Recommendation technologies have been born more than twenty years. Academic research and enterprise research give it a boost during the period. With massive information and distributed computation, big data-based recommendation emerged. Nowadays, recommendation module plays an important role in an electronic commerce system.This thesis starts with a review of recommender system. It involves types of recommender algorithms, including collaborative filtering and content-based recommendation etc. It describes the implementation of collaborative filtering based on Mahout and adopts Hadoop, Mahout to deal with course dataset. Then, it configures Sqoop for efficiently transferring bulk data between HDFS and MySQL to provide recommender information. Finally, use Maven, Tomcat and Spring to build a course recommender system based on Hadoop and Mahout.Contributions of this thesis are summed up in the following aspects.(1) Introduce the overall frame of Mahout, involving the implementation classes provided by Mahout in terms of recommendation.(2) Select recommender algorithms according to user requirements and characteristics of algorithms. Tuning parameters based on performances of recommender algorithms.(3) Implement a course recommender system based on Hadoop and Mahout.