【摘要】 由相互协作的用户在社会性标注系统中产生的大量的标注数据可以作为各种语义网应用的数据源.文中提出一种基于社会性标注的本体学习方法来挖掘蕴涵在社会性标注中的语义信息,提出一种隐含包容层次结构来刻画标签空间中潜在的结构,并基于此模型推导出本体学习算法.首先利用集合论的方法确定标签之间的包容关系,并将其表示为标签包容关系图.在将此图转化为层次关系时,为解决包容关系的不一致性,提出一种基于随机游走的标签普遍性排序方法.最后提出一种自顶向下的凝聚式层次聚类算法来生成概念层次结构.在实际社会性标注系统中采集的数据集上进行的实验表明,与目前的代表性方法相比,文中提出的方法在性能上有明显的提高.更多还原

【Abstract】 This paper proposes an ontology induction approach to harvest the emergent semantics from the folksonomies which are built from the social annotations made by collaborative users.The authors leverage a latent subsumption hierarchy model to uncover the implicit structure of tag space.First,tag subsumptions are identified with a set-theoretical approach and model the tag space as a tag subsumption graph.Then,a tag generality ranking procedure is used to overcome the problem of inconsistent subsumptions.Finally,an agglomerative hierarchical clustering algorithm is utilized to generate the concept hierarchy.The authors conducted experiments on a dataset collected from a real-world system.Both qualitative and quantitative experimental results show a competitive performance of the proposed approach.更多还原