Research Of Personalized Recommendation Systems For E-commerce Based On Complex Network Data Mining

Recommendation system is becoming effective method to solve the highly explosion of information on Internet, and is an important research content of E-Commerce area. Under the increasingly fierce competition, recommendation system can enhance E-Commerce sales by converting browsers into buyers, increasing cross-sells and building loyalty to prevent users losing effectively, which has good prospect for development.Various rule-based recommendation system, content-based recommendation system, collaborative filtering recommendation system and hybrid recommendation system which are currently extensive used are analyzed and compared in this paper; the four research content of the Recommendation Systems for E-Commerce, the application of the collaborative filtering algorithm in recommend system, the application of the complex network and the statistical characteristics in recommend system are studied in this paper. In the study, for the problems of data sparsity, cold-start and scalability in the collaborative filtering recommendation algorithm, the idea of the complex networks is introduced into the recommendation system, which ignore the items features of the users and commodity, they are viewed as abstract node, various statistical characteristics of the complex network is used to describe the various correlation between the users or commodities in the recommendation system. Concrete work as follows:The recommendation system based on the bipartite network is reappeared in Movielens data set, take the example of movie recommendation system, the users and the items are viewed as two kinds of nodes in the bipartite network, and how to deduce the recommendation algorithm according to the principle of resources distribution of the bipartite network. The experimental results show: the recommendation algorithm based on bipartite network beyond the classical collaborative filtering recommendation algorithm in accuracy, it is contributive to solve the problem of the cold-start, and be beneficial for the recommendation of the unsought goods.In this thesis, based on the complex network theory, a novel recommendation algorithm based on the user associated network is proposed. And also take the example of movie recommendation system to conduct experiments in Movielens data set. Firstly, the associated network between users is constructed according to the number of choices of the same movie and the assessment information, the extent of the connection between users is measured by value. Then, in the connection network between users that has been constructed, according to the purchase record of the commodity of all users, the commodity is recommended to these users who have a larger connection value. The recommendation of commodity items in the complex network is realized, in other words, the potential interested users of the commodity items is mined. The experimental results show that the recommended algorithm based on user association network, its recommentation accuracy is influenced by the connection weigth between users, in the recommended results that is produced, the links between users and movies that are presented are not exist in the original data, it is important to mine this kinds of links for data mining, even though the rate of accuracy is low, it is valueable to used in business applications.