英文摘要 | With the development of Web2.0, social tagging systems such as Flickr and CiteULike have gradually become a novel platform for knowledge sharing. Users are encouraged to annotate web resources (e.g., photos, papers, etc.) with freely chosen words, called tags, and they may also join groups to share relevant resources with users of common interests. Tags not only indicate users’ understanding of a resource, but also reflect the contents of the resource. The aggregation of tags constructs a high-level semantic description of a group. Therefore, tags perform as a bridge between users and resources, users and group, as well as resources and groups. Exploring relations among these entities (e.g. users, tags resources and groups) can improve information services in social tagging systems. The support of groups enhances the social aspect of tagging system. However, the huge volume of groups brings troubles for users to decide which group to choose. By mining the latent associations of users and groups through the bridge of tags, we can suggest potential groups to users. Tags and groups describe a resource from different angles, meanwhile, semantic tags and similar groups may help users’ better understand a resource. By integrating all such information, we can refine the joint tag and group recommendation, which is good for resource discovering and knowledge spreading. Nowadays, resources in a group always ordered by its sharing time; in this manner, a large number of high-quality resources will disappear from main page as time goes by. As groups aggregate tagging behaviors, the co-occurrence of tags provides a possible way for topic-oriented resource browsing. Users’ tagging behaviors are time sensitive and interest drifts exist in social tagging systems. Therefore, incorporating interest drifts to discover users’ current preferences may provide more accurate information services. Based on the photo tagging system Flickr and paper tagging system CiteULike, the main focuses of this thesis are summarized as follows: 1. We propose a tensor decomposition based group recommendation approach to combine tag related multi-mode entities. Firstly, a three-mode tensor is constructed from “user-tag-group”usage data. Then we discover the latent topics from the three-mode tensor based on non-negative tensor decomposition. Finally, according to the latent associations between users and unjoined groups through latent topics, groups will be suggested to users. Empirical... |
修改评论