Technology Research And System Design Of Personalized Recommendation

With the rapid development of e-commerce and Internet technology, the number of informationresources in network is exponential. Vast amounts of information appear in front of the user, whichmakes users feel at a loss, and they are difficult to find the interesting resources, thus it leads to theso-called "information overload" problem. Personalized recommendation system makes thisproblem eased, and recommends more targeted information to users through the study of userbehavior, interests, and the environment. In order to study personalized recommendation systembetter, this thesis focuses research in the following areas：1. This thesis analyzes the main features and data acquisition modes of social networks.According to the nature and classification of the trust, we give a measure of trust, and propose aweighted calculation method based on similarity and trust. Meanwhile, several classic communitydiscovery algorithms are introduced and compared with the advantages and disadvantages, then animproved method is proposed and used in personalized recommendation. Finally, experiments aremade to compare the performance under different conditions, and to verify the effectiveness of thepersonalized recommendation system.2. This thesis summarizes the definition and research status of context-aware, and elaborates theway of acquisition mode: explicitly, implicitly, inferring, and describes the recommend ways ofcontext information: pre-filtration, post-filtration, modeling. Based on these theories, we proposes abased on context-aware integration personalized recommendation algorithm. The algorithm dividescontext into physical context and user preferences context. A Bayesian network constituted by time,place, and other situational factors is built to match the physical context. Meanwhile, the timewindow and content-based recommendation algorithm are combined to match preference context.Then, we weight matching and recommend information resources to the user. Finally, a commondata set and the evaluation are used to make experiments, and the results indicate that the algorithmimproves the accuracy of recommendation.3. This thesis introduces several commonly used personalized recommendation technologies,and gives a comparative analysis of the advantages and disadvantages, and describes the variousapplications. On the basis of two algorithms, a personalized recommendation system is designed,which describes the overall framework and the various modules, and analyzes the functionality andworkflow. This system constitutes the client and the server. The client implements on Android platform, and it includes sharing, query, location-based services and other modules. The serverincludes personalized recommendations, database, network communications modules.