Research On Semantic Analysis Of Tags Based On Link Analysis And Clustering Methods

The development of social tagging system provides a novel way to organize the information on the Internet. Inheriting the characteristics that treating the users as the core, social tagging systems make fully use of the knowledge and expertise of the ordinary users. Currently, an increasing number of information systems support the functions of social tagging. From both perspectives of the diversity of supported resource and the roughness of provided functions, the continually improving of social tagging system servers the Internet users a more convenient way to manage, share and retrieve the information.Within the social tagging system, users can annotate the cared resources with tags in an unrestricted way. As the most significant character of the social tagging, tags are the explicit semantic descriptors of the content of the resources as well as the implicit reflection upon the preferences of the users. The freedom use of tags, which makes users manage the interested resources without any restriction, allows the users from any filed participating into the construction of folksonomy.However, like a double-edge sword, the free use of tags also brings some problems on social tagging analysis. First of all, the ambiguity hinders the recommendation, classification and retrieval. Also, the huge data space impedes the fast and accurate analysis of social tagging. Commonly, tag exists as the form of words or phrases, but they co-annotate some resources or are co-used by some users, which provides a way to discover the indirect relationship among tags. Hence, the thesis starts from these relationships and further analyses the problems caused by tags.Firstly, basic introductions of social tagging systems are given in the first part, such as the way of system modelling, characteristics of the system and the problems confronted. Then, based on the introduced problems, link analysis and clustering-based methods are provides. For the link analysis method, users are marked in order to find the exemplary users and preventative tagging behaviors. For the clustering method, the topic-based user model schedule is given firstly, and the proposed algorithm based on the latent semantic topic is introduced. The last part of the thesis evaluates the performances of the proposed two methods with the real datasets of Delicious and Movielens.