Research Of Recommendation Technology Based On Social Tagging System

With the overload informations of Web2.0 web application, the social tagging system is becoming increasingly popular because of low threshold, flexible operation, ease of use and other characteristics. In social tagging system, users can label resources with tags. The tags are any keywords chose by users to meet their own preferences. Social tags can help users organize, share and discover information resources more effectively, and deliver the user's interests of the general public. At present, how to recommend tags for users accurately and effectively, and how to recommend resources for users based the tags information have become important research directions of recommendation field.Our research work is based on social tagging system, the main work include the following two points:1. Presents a tag recommendation method to meet the personalized features of users'. In social tagging systems, because of the freedom of tags and the various preferences of different users, every user will have a tags set with its own characteristics. Aim at the problem of personalized tags; this paper put forward the tag recommend method of combining transfer matrix method with collaborative filtering. With the public wisdom, to recommend the target users tags which can meet their individual needs. Transition matrix is the mapping between users'personal tag vocabularies and the corresponding folksonomies, which established based on the concurrence of tags. Prior to generating the recommended tags, the tags'distribution of resources is mapped to the target user's label collection by the transfer matrix. This method can produce a collection of tags meeting the characteristics of the target user's personalized tags. But because considering the only target user's interest model, the target user can not be recommended with new tag. Therefore, we fuse the collaborative filtering with the transfer matrix method, depending on the target user and his nearest neighbors' interest models to produce the final recommendation tags. Not only to meet the user's personalized features, but also can discover new interested tags for users.2. Apply the LDA model to information recommendation in the social tagging system. The tags in social tagging system not only indicate the features of resources, but also represent the interests of users', so they provide important informations for information recommendation. However, due to the open of social tagging system, there are redundant tags, ambiguity tags and garbage tags problems. To solve these problems, this paper utilize LDA model, to find the latent semantic themes in tagging system. And then to find the resources meet the interest of every user on the semantic level best. And to determine nearest neighbors for users based on semantic models. With the collaborative filtering ideas, recommend resources based on target user and their neighbors' interest model. Compared with the methods only based on tags space, this method can eliminate the semantic ambiguity problems such as synonymy tags, polysemy tags and so on.Experiments on the delicious corpus prove that, the tag recommendation method combining transfer matrix with collaborative filtering, and the resource recommendation method based on LDA model in social tagging system can produce better recommendation results.