Abstract: The flexibility of Collaborative Tagging Systems（CTS） brings large number of synonymous and polysemous tags which make the use of these profile information to personalize resource recommendation difficult.Graph-based tag clustering is proposed to form groups of semantically-related tags.Then the tag clusters act as an intermediary between users and resource and are utilized to personalize the query results in CTS.5-fold cross-validation is performed on two data sets，and the results are compared with two other algorithms.Results show that the proposed algorithm demonstrate much better personalization measured by the F-value，and the effect is more miraculous in a multi-topic than in a single-topic CTS.This observation suggests that in a multi-topic CTS tag clustering such as proposed in this paper is an important strategy.

Key words: collaborative tagging, recommendation, graph clustering, personalization

摘要： 协同标记系统允许用户自由标记系统资源，但也由此产生了同义标签和多义标签问题，这使得如何利用用户标签构成的用户概貌信息进行个性化资源推荐成为一个难题。为此首先基于图聚类算法把系统中语义相近的标签构成聚类，然后以标签聚类为中介衡量特定用户和资源的相关度。在BibSonomy和Delicious两个数据集上进行了测试，并和另外两种算法进行了对比。实验结果显示应用该算法产生的推荐，其性能优于对比算法，在主题宽泛的系统中效果尤为明显。说明协同标记系统首先进行标签聚类是产生个性化资源推荐的重要方法。

关键词: 协同标记系统, 推荐, 图聚类, 个性化