Dissertation > Excellent graduate degree dissertation topics show

Research on Personalized Recommender Algorithm in E-Commerce Based on Collaborative Filtering

Author: LiYaXin
Tutor: MaGang
School: Dongbei University of Finance
Course: E-commerce
Keywords: E-commerce personalized recommendation system Collaborative filtering Singular Value Decomposition BP neural network
CLC: F713.36
Type: Master's thesis
Year: 2010
Downloads: 459
Quote: 0
Read: Download Dissertation

Abstract

With the popularity of the Internet and e-commerce applications, extensive, supply chain and logistics continues to improve, people enjoy the convenience of online shopping is also caught in a dilemma of information overload, a large number of users is difficult to find the product information they need goods. Thus, e-commerce recommendation system came into being. Recommended System in e-commerce platform to play the role of sales staff, recommended to the user of goods they need to successfully complete the purchase process. E-commerce recommendation system has good prospects for the development and application has gradually become an important research content, personalized recommendation system is designed to meet the specific needs of different users to produce, and it is one of the research branch. Collaborative filtering recommendation algorithm is currently recommended system most widely recommended techniques, in many respects than other recommendation algorithm shows outstanding advantages. However, collaborative filtering recommendation algorithm is not mature enough, there are still some problems, such as data sparsity problem, the system scalability issues and the time factor issues. In this paper, e-commerce recommendation systems and collaborative filtering recommendation algorithm the main problems facing the recommendation system recommended strategies and key technologies such recommendation algorithm for a useful exploration and research. The research work is mainly from the following four aspects: First, the introduction of existing e-commerce recommendation system and the main recommendation technology. E-commerce personalized recommendation system and the main recommended techniques to re-sort and summarize the relevant literature, draw commerce personalized recommendation system research content and composition analysis of various major recommendation technology implementation process, advantages and disadvantages, and then analyzes each recommended techniques applicable. Secondly, the collaborative filtering recommendation algorithm, especially the traditional collaborative filtering recommendation algorithm research. Here on collaborative filtering recommendation algorithm related literature, re-define collaborative filtering recommendation, noting that the principle of collaborative filtering recommendation algorithm and implementation process. And focuses on the traditional collaborative filtering recommendation algorithm, analyze its strengths and weaknesses. On this basis, summed collaborative filtering recommendation algorithm facing bottlenecks and solutions. Again, collaborative filtering recommendation algorithm is a core part of this article. Collaborative filtering recommendation algorithm for the existence of data sparsity and the time factor is proposed based on SVD and time-weighted algorithm improvements. The algorithm is based on a literature, joined the singular value decomposition method and time functions, and the similarity measure to improve the underlying similarity measure formula. The first step, using the singular value decomposition method for the user - project evaluation matrix dimension reduction, reducing data sparsity. The second step, using BP neural network to populate the dimensionality reduction users - project evaluation matrix ungraded items, once again reduce the data sparsity. The third step, using the calculated correlation similarity measure the similarity between users, the user of the nearest neighbor set. The final step in the forecast period taking into account the recommendation recommendation system for the timeliness, based on the recommendation of the time-weighted prediction formula, closer to the time of the evaluation data to give greater weight to the evaluation time data given far less weight, improve the prediction accuracy. On the whole, to a certain extent, solve the data sparseness problem and the time factor to improve the quality of e-commerce recommendation system. Finally, the simulation test improved algorithm. Use the Matlab software for the Minnesota State University GroupLens research team provides common data set MovieLens simulation testing to verify its legitimacy and effectiveness. In this paper the design of two test programs are different levels and different data sparse number of nearest neighbors, have proved that the improved algorithm is better than the original algorithm has obvious advantages, improve the quality of e-commerce recommendation system. Based on the above research, innovation of this paper is as follows: First, the concept of collaborative filtering to give re-definition. Second, improve the algorithm, the SVD and BP neural network combination, dual reduces the user - project evaluation matrix sparsity. Third, taking into account the evaluation of real-time information, the application of time-based function prediction formula recommended. Fourth, the design simulation test proved that the improved algorithm can effectively improve the quality of recommendation system.

Related Dissertations

CLC: > Economic > Trade and Economic > Domestic Trade and Economic > The circulation of commodities and the market > Sale of goods > E-commerce,online trading
© 2012 www.DissertationTopic.Net Mobile