The Study Of Social Tagging Model Based On Tag Popularity

With the development of Web2.0, a lot of social tagging systems such as Del.icio.us, Last.fm and Flickr emerge and foloknsonomies based on these social-tagging-systems prosper. The study on the emergence of folksonomy has aroused widespread concern in academic circles, especially in information science, computer science and the field of complex systems. The formation and intrinsic tagging mechanism about folksonomies have been widely discussed in scientific community. Social tagging is essentially a process in which knowledge is collectively generated from individual tagging activities.In this sense, this process is a means of collective intelligence. This thesis investigates the dynamics in social tagging system by using the agent-based modeling method, so as to explore the underlying collective intelligence.The users’behaviors in social tagging systems determine the properties of folksonomies, so it is critical to understand the regularity of user behaviors. Different from the mainstream of existing tagging-dynamics models, which are commonly based the co-occurrence of tags, this thesis analyzes the characteristics of users’behaviors by investigating the characteristics of tag-resource distribution. We proposed a multi-agent model from two main aspects:users’ background knowledge and system recommendation. Using this model, we conduct multiple sets of fitting experiments on the tag-resource distributions in real tagging systems. The results shows that the model presented in this thesis can effectively reflect the characteristics of the tag-resource streams in the real datasets and further prove the analysis of users’tagging behaviors reasonable. The innovation of this study can be mainly summarized as follows.(1) Starting from the analysis of the posts in Del.icio.us, we observed that tag-resource distribution presents different slopes of power-law distribution which is different from what the previous studies mentioned, namely segmented power law characteristics and approximate "platform" structural features. We speculate that system recommendation mechanism may account for the generation of tags on the platform.(2) In the previous work, it is generally assumed that users add a tag to a resource from the whole characteristics of it, in this thesis, however, we suggest that users tag a resource from one dimension of the resource attributes and each resource has multi-dimensional attributes to identify the type of resource characteristics. Results of agent-based simulations reveal that the tagging behavior based on users’divergent knowledge background may be the main cause to generate the segmented power-law characteristic of the tag distributions. (3) In model validation, first of all, unlike the previous studies in which researches evaluate the model from the co-occurrence frequency distribution of the tags as well as the growth of different tags, we use tag-resource distribution to evaluate the model present in this thesis and obtain a better fitting results with the real datasets; Secondly, we use the real user annotation data from three different social tagging systems, namely Del.icio.us, Last.fm and Flickr to evaluate the model, showing that the model has good adaptability and scalability.In all this thesis presents a model for deeply understanding the emergence of collective intelligence in social tagging. The study not only has important theoretical significance, but also has a strong practical significance for the general public to construct a reasonable folksonomy and improving retrieval efficiency of social navigation, and a better user interface design.