The Research Of A Tag-based Personalized Recommendation System

Recent years, with the development of Web2.0 and social networking sites, the Internet has changed dramatically, individuals convert from the reader of information into a publisher. In this case, the amount of information grows much faster and the site's recommendation system faces enormous challenges. The tag system as the basis function of Web2.0 can reflect the user interests more comprehensively, has been used by more and more sites. Therefore, as the new recommendation system currently, the tag-based recommendation system can acquire a result of recommendation. The result is much closer to the user needs and has a high research value.This thesis studied the user profile representation and tag technology, understanded the user's interests according the tag which user added and created a user model. The thesis also pointed out the defect of the traditional tag weight calculation method and given an improved label weight calculation algorithm. The weight calculated by the new algorithm can reflect the user's interest to the recourse more accurately. Then we combined the weight with tag time information, given a user profile update algorithm. Furthermore, we can predict the user interests from the viewpoint of other groups, expanding the range of recommendation. Thus the thesis designed four mode of user profile-- personal, social, friends and global, to meets the user's individual needs. Different from the user to resource scored matrix of traditional collaborative filtering recommendation system, this thesis defines a tag to tag matrix to reflect the relationship between user and resources. Based on the matrix, a tag-based recommendation algorithm and corresponding test results were given by the thesis. Finally, we designed and implemented a tag-based personalized recommendation system, analyzed the system results and functional modules from the point of system design and given function test results. Research and experimental results show that this thesis using tag as the source of user interests, could acquire more comprehensive and accurate user profile. And based on this, recommendation algorithm can improve the system's hit rate significantly and the recommendation system can meets users'individual needs better.