Research Of Information Recommendation Algorithm Based On Bipartite Network

With the rapid development of network and computer technology, there arehuge amounts of complex information on the Internet, and it’s so hard for people toobtain the required information or products quickly and efficiently, and then person-alized recommendation systems come into being. Personalized recommendation sys-tems collect user information, perform data analysis and modeling, predict users’potential needs and then recommend personalized products for users. Personalizedrecommendation system can also help the sellers to find what the user’s interests are,so that they know how they can attract the customers and then boost sales.A personalized recommendation system is mainly composed of input module,recommendation algorithm module and output module. Wherein, the recommendati-on algorithm module is the core part of recommendation system. How to analyzedata efficiently and recommend products to the target users becomes the top priorityof the personalized recommendation system.In this paper, user similarity method of collaborative filtering is introduced tohybrid recommendation based on heat conduction and mass diffusion, and theproduct similarity method is proposed, which is evaluated by using the benchmarkdatabase called MovieLens. Experimental results show that the accuracy, popularityand diversity of the improved algorithm are greatly increased.In addition, statistic analysis on the data sets is performed, and the result showsthat the interest of a majority of users is very diferent from the popularity of products.Thus the factor of user interest is introduced into the mass diffusion algorithm, and theproducts which are closer to the interests of user will be given higher weights in theprocess of resource redistribution. Numerical result shows that, compared with themass diffusion, the algorithm based on user interest can achieve better accuracy,popularity and diversity.