Research And Implementation Of Recommendation System Based On Mahout

With the rapid development of information technology, the information explosion does not make the utilization of information increase; on contrary it reduces the utilization of information instead. How to find the information we truly need from massive amounts of data will become increasingly difficult. In order to solve the problem of information overload, academia and industry already carried out a variety of exploration. The recommendation system is a new information mining mechanism which has been widely used in many fields. Besides, with the increasing amount of data, recommendation system is facing the challenge of scalability of analysis, storage space of scalability and so on.Mahout is a machine learning algorithm library which designed to provide programmers with efficient algorithms examples. Hadoop is a distributed computing and storage platform. Those emerging technologies provide a new way for the design of recommendation system. In the thesis, based on Mahout and Hadoop, we researched on the recommendation systems and distributed parallel computing, carried out researches in the following four aspects:we analyzed the classical algorithm of recommend field, related technologies, and expounded their respective applications. Then we designed and implemented a recommendation system based on Mahout "taste" engine. Because of the similarity calculation is largest in the recommendation module. In order to speed up computation, we introduced the MapReduce programming framework to offline parallel calculate similarity. And we used HDFS API to implement storage of user data, which can improve the scalability of the system memory space. Finally we studied the evaluation methods, designed and implemented the evaluation modules to evaluate various parameters, and used JFreeChart Java graphics library to visualize the results, which can help researcher flexibly analyze of data and select appropriate algorithms and parameters easily.