Collaborative Filtering Recommendation Algorithm Combining Trust Mechanism With User Preferences

With the growing popularity of online shopping, recommendation system not onlybecame the business tool to improve product sales of e-commerce enterprise, but alsoincreasingly important decision tool of many users. Internet opens a large of productsand advertisings to users. It is a waste of time and trouble thing to choose from manyproducts, which guides the development of the recommendation system. The typicaloperating of recommendation system is to recommend the most suitable products tousers by analyzing the users’ interest model. In the area of electronic commerce, use thepersonalization technology for deliver products to users can alleviate the problem ofinformation overload. Other, it can reduce the customer’s workload and increase theprobability of customer’s loyalty to enterprise. Researchers and industry practitionersare looking for all ways to improve the performance of recommendation. Even a smallimprovement may also bring a big commercial return. Traditional recommendationsystem application technology includes recommendation based on content、recommendation based on collaborative filtering and so on. At present, collaborativefiltering is the most successful recommendation technique. It has been widely used inpractical applications, such as recommend music, books and files.In this thesis, the focus of research work and results are as follows:①Mutiaspect deep-going study on recommendation. Analyze its classification,function, organization framework, module division and how to use the recommendationresults etc… Introduce several popular recommendation technologies in details,advantages and disadvantages of their own. Introduce hybrid technology which is orderto make up a single recommendation technology.②Do the collaborative filtering as the center and do a full range of analysis on it.The traditional collaborative filtering exists a series of problems, such as data sparse, anew user. The paper analyzes the reasons and consequences. Take these problems as thestart pointing and put forward the solution.③Analyze the shortage of rate. Rate can only express whether the user is satisfiedwith the projects or not. It isn’t able to show the user’s intrinsic interest on the project.In order to deeply mining the user’s real interest on the project, put forward the conceptof user interest degree.④Traditional collaborative filtering algorithm rely solely on the similarity to locate neighbor users, which has a serious negative impact on the performance of. Importingthe trust mechanism in social networks, A model is build up from the perspectiveof individual’s subjective trust and global reputation in social circle. Firstly, direct trustis formed by considering users’ interactions, rating differences and preferences.Secondly, indirect trust can be generated by reputation and expert prior trust model.Then the proposed algorithm constructs the trust relationship among users through adynamic weighting manner. Finally, the parameter η is to coordinate double attributesincluding trust and similarity to make the user relationship closer, which can solve theproblems of new users as well as sparsity effectively. Pragmatic analysis reveals that theimproved model has the remarkable results.⑤Take the MovieLens dataset as experimental subjects and achieve the improvedalgorithm. Then compare the experimental result with traditional algorithms and avariety of related algorithms.