A Study On Collaborative Filtering Recommendation Technology Based On Users' Browsing Paths

In recent years, with the fierce development of the e-commerce, it brings revolutionary changes to the commodity trading. Products are divided by electronic commerce websites on client customer group. E-commerce websites give service around clients and offer them goods which they needed. So it is necessary to provide individuation service for every user. So Personalized Recommendation system also has become a hot technology at present. Personalized Recommendation is a system based on user browsing path of Mining Technology Recommended system, with the aim of the website user-friendly visited. It can predict the future and the number of users-loving, as well as e-commerce enterprises to provide basis for decision-making. How to combining browsing path Mining Technology and recommended Technology is an important issue of personalized recommendation system. With this issue, the following work has done in this paper:Firstly, this paper choice collaborative filters technology as a recommended study, and introduces its current research status, problems and basic theory. Secondly, this paper introduces individuation recommendation system and its mining process of users' browsing fondness paths mainly. The mining algorithm of users' browsing fondness paths is given. Thirdly, this paper shows collaborative filtering recommendation method based on users' browsing fondness paths. It is innovation of the article. Hidden Markov models provide an effective method for implementing collaborative filtering recommendation method based on users' browsing fondness paths. HMM simulates the behavior based on users' browsing paths when they browse web site and set up nearest-neighbor set of browsing paths. Used HMM instead of similitude model to measure users' similarity, the nicety of nearest-neighbor commendation is improved greatly. And it settled real-time problem and the extension of data room. Then fancy degree is set up. As adding fancy degree of users', the item which is commended for target user is more suitable. Combined the fancy degree, the prediction model of dynamic collaboration filtering recommendation is given. Finally, in order to give better satisfaction neediness for users and improve recommendation nicety, the HMM is updated, various users' data is dealt effetely. For introducing and comparing to HMM, dynamic Bayesian network is the first choice as the update model. As the flexibility modeling of DBN, it is used in fusion algorithm. To implement update learning of network structure, that is, join new character into HMM commendation model to construct DBN commendation model in order to updating commendation model. For combining all old and new samples to learn and train, time is saved. And network structure is optimized. The commendation model is satisfying users' neediness better.This article describes mainly the combination of users' browsing paths and collaboration filtering recommendation. As users' browsing paths used as basis of commending data, the recommendation link is given immediately while users are browsing pages. And the corresponding recommendation result is given according to the changes of users' fondness. Combine HMM and DBN to set up a new users' forecast model. This model joins the training result of the HMM and DBN. And dynamic changes of users' fondness are reflected. Finally, it points out the deficiency, and looks forward to the direction that can be studied further in the future.