Study Of Recommendation System Based On Collaborative Filtering Algorithm

With the popularization of Internet and the rapid development of E-commerce, many famous E-Commerce sites have developed Recommender system for providing personalization service to consumers. Recommender systems are used by E-commerce sites to suggest products their customers and to provide consumers with information to help them decide which products to purchase. This article researched the personalized recommender system and its main recommendation technologies, especailly collaborative filtering algorithm including User-based collaborative algorithm and Item-based collaborative algorithm. The expansion and redundance of information bring forth people the puzzle of choosing information.Collaborative filtering algorithm has been very successful in both research and practice, especially in the effect and precision of recommendation. But it suffer from sparsity and cold-start problem which affect the performance of recommendation system badly, which could not take advantage of the Collaborative filtering.Based on the analysis of these, the detail works is shown below:1.Decomposition the user-item rating matrix, and validate that this method could reduce the noise of matrix and can open out the latent relevancy between users ,solving sparsity problem.2.Computing the similarity between users, our experiment validate that the measurement of Correlation-based Similarity has more higher quality of recommendation than that of Cosine-based Similarity.3.At the step of prediction, our experiment use the method of Deviation-from-Mean to predict new item ratings that users have not rated and validate that this method has higher quality of recommendation than that of user-based and Item-based collaborative filtering algorithm. Meanwhile, it remedy the disadvantage of new-item in cold startup problem, increasing the quality and precision of recommendation.