Research On Personalized Recommender Algorithm In E-commerce Based On Collaborative Filtering

With the popularity of Internet and the extensive application of E-Commerce, also the continuous improvement of supply chain and modern logistics, people enjoy the convenience of shopping online. But there are problems come behind. Such as, information overload which make people can't find their target products. Therefore, E-Commerce Recommendation System comes into being. In e-commerce platform, Recommendation System works as a sales staff, referring users to the products they need and complete the purchase process successfully. It has a good development and application prospective and become an important research gradually. Personalized Recommendation System which is designed to meet different users'specific needs also is one of the study branches.Collaborative filtering recommendation algorithm is the most widely used recommending technique, and show its outstanding advantages compared to other recommend algorithm in many aspects. But it also has many problems, such as data sparseness problem, the system scalability problem and time factor problem. This paper does a useful exploration and researches on dealing with the strategies and key technologies which collaborative filtering recommendation algorithm and e-commence recommendation system have.The main contents of this paper are divided into the following four areas.First, the research on the existing e-commerce recommendation system and the main recommendation technologies. Collate and summarize the relevant literature on e-commence personalized recommendation system and main recommendation technology, obtain its research content and composition, then analyze the implementation process and the advantages and the disadvantages of the main recommendation technologies, finally analyze the application of each recommendation technique.Second, the research on collaborative filtering recommendation technology. Redefine the collaborative filtering and indicate its principle and implementation process. The stress of this part is to discuss traditional collaborative filtering algorithms and to analyze the merits and drawbacks. On base of above, indicate collaborative filtering algorithms'problems and solutions.Third, put forward improved algorithm based on SVD and time function. The improved algorithm is based on a literature, joined the singular value decomposition method and time functions, and similarity measures to improve the formula for the correlation similarity measure. Firstly, apply singular value decomposition method to the User-Project Evaluation Matrix, and recent the matrix's dimension, then reducing the degree of data sparseness. Secondly, use BP neural network to fill the no-scored items of the matrix, reducing the degree of data sparseness again. Thirdly, using correlation similarity measure calculates the similarity between users and the user's set of nearest neighbor. Finally, consider the timeliness of the recommendation system in the stage of recommending prediction, and use time-weighted based on the recommendation of prediction equations, give greater weight to the more recent data and give far less weight to the less recent data, improve the forecasting accuracy.On the whole, the improved algorithm solve the sparse data problem and the time factor to a contain extent and improve the recommendation quality of e-commerce personalized recommendation system.Finally, the simulation test of improved algorithm. For the improved algorithm based on SVD and the time-weighted, using Matlab software on the MovieLens data set, which is offered by the GroupLens in the University of Minnesota research team to verify its legitimacy and effectiveness by simulation testing. Both the two test programs of this paper can prove that the improved algorithm is better than the original algorithm obviously, and improve the quality of e-commerce recommendation system recommended.Based on the above research, the innovations of this paper are as follows.Firstly, refine the concept of the collaborative filtering.Secondly, combining SVD and BP neural network measures to reduce the User-Project Evaluation Matrix's sparseness doublely.Thirdly, consider the timeliness of the recommendation system in the stage of recommending prediction.Finally, design the simulation experiments to demonstrate that the improved algorithm can improve the quality of the recommendation system.