标题: | 以个人化标签推荐系统探讨网路标签使用行为 Investigating user tagging behaviors in social bookmark system |
作者: | 邓睿清 Jui-Ching Teng 孙春在 Chuen-Tsat Sun 资讯学院资讯学程 |
关键字: | 群众分类;社会标签;社会书签;标签系统;推荐系统;folksonomy;social tagging;social bookmarking;tagging system;recommendation system;clique filtering |
公开日期: | 2007 |
摘要: | 随着web2.0 时代的来临,网路应用掀起一股tagging 风潮。资讯的消费者将资讯贴上标签不只是管理个人的档案,也管理个人的知识。同时藉由网路的分享,共享标签更是让知识传递及汇集的速度到达前所未有的境界。标签推荐可以提供候选字以及标签关联性使得知识管理、资讯取回、搜寻排序更有效率。 在观察社会标签系统(social tagging system)中由使用者、标签、文件所形成的三分关联网路(tripartite network)之后认为,社会标签系统中一个好的推荐系统应展现使用者们所形成的群体智慧,以众人的智慧帮助个人,也以个人的力量帮助众人。在过去的标签推荐演算法在寻找相似人群以及相似标签上的局限,本论文提出一个基于社会网路分析理论的使用者标签推荐的演算方法“派系筛检法(Clique Fitering)”,演算法的架构主要是衍生于协同过滤法(Collaborative Filtering)但其中演算的精神来自于社会网路中的派系过滤法(Clique Percolation),在使用者对文件贴标签或使用者想利用标签对文件进行过滤的情境(scenario)下提高标签推荐的准确度。以这种演算法,可以直接应用在标签推荐系统中,不需系统对于字词有所认知,可以适用于现行以社会标签为管理的应用系统中, 并且也可以将结果应用在其他个人化的系统当中。除此之外,也以目前最多人使用的论文文献检索系统CiteULike 作为范例,利用社会网路的分析方法,分析其中人际之间标签使用的群聚行为,发现标签的使用反映人的思考,标签使用的习惯有“物以类群、人以群分”的小世界行为模式。 Folksonomy is a popular application of web2.0. Information consumers label resources with arbitrary words, so-called tags. Social tagging systems not only help people share resources but also share knowledge. Tag recommendations can help user in Knowledge Management by providing candidate tags, in Information Retrieval by discovering relations of tags, and in Search by providing personalized keywords reminding. After observing social tagging system, we focus on the tripartite network that formed by users, tags, and items in the system. A good recommend system should present the co-active intelligence. In the past, tag recommendation algorithm is difficult to find similar people and similar tags. In order to improve tag recommendation, we propose a modified tag recommendation approach Clique-Filtering that based on social network theory. We evaluate and compare it with collaborative filtering on real-life dataset. We show the performance of modified approach is better than collaborative filtering in the sparse dataset. We can apply the result to other personalized recommendation system. After analyzing one of popular bibliography site CiteULike, we discover the personal tag clustering in the real-life system. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT009567572 http://hdl.handle.net/11536/39861 |
显示于类别: | Thesis |
文件中的档案:
If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.