Research Of Personalized Recommendation Technology Based On The Graph Model

E-commerce has become an important means of trade liberalization and economic globalization, and is also the key driver of traditional industries, E-commerce can save costs companies using to do commercial activities, streamline business processes and reduce its operating cycle, thereby increasing economic efficiency. Meanwhile, the rapid development of electronic commerce makes information overload. It is difficult for users to find the products they want, and companies can not make the right decisions, the resulting e-commerce recommendation system. But the existing e-commerce recommendation systems have many shortcomings, such as data sparseness reducing recommendation quality, the poor system scalability, useful information expressed incompletely, and so on. Paper solves those problems, explores and researches personalized recommendation technology by graph model. The main contents and findings are shown in the following.First, we introduce the concepts and functions of e-commerce and personalized recommendation system, describe the composition of personalized recommendation system and its related technologies. We also analyze the construction process and advantages of the graph model, and describe the resources and characteristics of the small world network.Secondly, we construct a user-layer network using the small world model, introduce the weighted edges into the traditional small world network, and measure weights based on the similarity. In user network, we define flow resistance and the smallest flow-resistance that is used to measure the differences between user nodes, and by which we define flow-efficiency of the network, which unifies the measure of global and local clustering features.Thirdly, we propose a edge-reconnect algorithm to build a weighted small world network, at the same time, we have the users on the user network clustering. Experiment results show the feasibility of the model and algorithm, and this method can improve the effect of user clustering.Finally, we build bayesian network in the product layer by structure and parameters learning, then we construct a graph model combining both user layer and product layer, and raise a personalized recommendation algorithm based on the graph model. Finally, we verify the feasibility and superiority.