The Research On Key Technologies Of Cloud-Based Personalized Recommendation System

The rapid development of Internet technology makes the scale of Internet ever-increasing, the amount of information is explosive growth. Faced with massive information, it's hard to find interested information resources without suitable tool. Recommendation system mines and analyses users' historical data, obtain user's preferences, and then recommend their favorite product or information resources to them to meet the massive users' personalized demand.With the increase of data processing capacity, scalability problem becomes the bottleneck of the development of recommendation system. The combination of recommendation system and Hadoop distributed cloud computing platform can solve this problem well.First of all, from the two aspects about HDFS and Map Reduce, this paper researches the operating mechanism and programming principle of Hadoop. Based on this, With the universal model of recommendation system,it designs a recommendation system architecture based on Hadoop, and introduces the design idea of each module in detail.Secondly, it analyzes the popular collaborative filtering algorithm and Slope One algorithm in detail, uses Mahout to achieve collaborative filtering algorithm on Hadoop platform, and designes Map Reduce parallel processing scheme of Slope One algorithm based on Hadoop.Once more, based on the traditional Slope One algorithm, this paper combines the idea of collaborative filtering, proposes KS-Slope One algorithm. The cosine similarity of the target user's K neighbor users is used as the weight of the difference between the items to improve the scalability of the algorithm, and this paper achieves KS-Slope One algorithm of Map Reduce parallel processing on Hadoop platform, it improves the data processing capability of the KS-Slope One algorithm.Then, the performance of KS-Slope One algorithm is tested by Movielens data set on Hadoop cluster. The experimental results show that the accuracy of KS-Slope One algorithm is significantly improved compared with the traditional Slope One algorithm and weighted Slope One algorithm, it proves the validity and advanced nature of the KS-Slope One algorithm.Finally, this paper uses the information and data in Movie Lens data set to implement two kinds of collaborative filtering algorithm and KS-Slope One algorithm, and shows the recommendation results.