The Research Of Personalized Recommendation Technology Based On Tag

With the rapid development of the Internet and Web2.0technology, the websitesprovide users with more and more information, but the network structure has becomeincreasingly complex. Among them, the most significant is the Social Tags which havebeen widely used in a great number of personalized websites. Most websites in Web2.0have a Collaborative Tagging System, such as Delicious and Flickr. There are more userinteractions in these websites, so the ways of obtaining information are different fromthe traditional ways that user obtain information through the browser. In the SocialTagging Systems, tags are used to organize and manage the resources by users, so tagsare represented as a bridge connecting users and resources. In addition, users not onlyare the consumers of the information, but also the provider of the information, this maycause serious problems which is called "information explosion" and "informationoverload". In order to solve such problems, the personalized recommendationtechnology comes into being.Because Collaborative Filtering Recommendation technology has good algorithmideas and perfect recommendation effect, it has been widely used by the conventionalrecommendation system. It mainly studies the users’ rating data to the resources, andthe research object is a two-dimensional structure containing users, resources. But in theSocial Tagging System, for the three-dimensional structure of users, tags, and resources,the conventional recommendation algorithm can not meet their needs, because thetraditional recommendation algorithms does not take the personalized information intags into account.As regard to the traditional personalized recommendation methods fail to solve theproblem of the demand for resource recommendation, this thesis proposes acollaborative filtering method based on tag and time factors. This method can improvethe traditional collaborative filtering methods, and can fully consider the characteristicsthat user’s interests may change over time. According to the tag information added byusers, we are able to analyze the use of the tag and can make the resource characteristicmore accurate. Besides, it can truly reflect the preferences of the individuals, forexample, the tags applied by users in the recent can reflect the recent interests of users.On the basis of this idea, we propose a user interest model method by using taginformation and the context of tagging behaviors. By analyzing tag information, tagging time and the rating data of resources, we can construct a pseudo rating matrix. Drawingon the experience of collaborative filtering algorithm, we can calculate the similaritiesof users and get the Nearest-neighbor users for a active user. In the end, the predictionresults are generated on the basis of the Nearest-neighbor users.By experiments, I verify the proposed recommendation method. The experimentalresults indicate that, compared with other traditional recommendation method, therecommendation effect of the proposed recommendation method is improved.