基于层次概念格的分面导航.docVIP

PAGE PAGE 1 基于层次概念格的分面导航* *何超，1982年生，男，江苏人，博士生。 何超1,2，程学旗1，郭嘉丰1 1中国科学院计算技术研究所，北京，100190 2中国科学院研究生院，北京，100190 E-mail: hechao@ 摘 要：分面导航是用户基于多维分类目录检索和浏览资源的主要方式之一。通过推荐与当前搜索结果相关的类别，帮助用户理解搜索结果，并有效避免查询结果为空。然而，目前的分面导航难以分析所推荐类别之间的深层语义。本文提出了一种层次概念格作为资源集的本体，它完整并简洁地描述查询结果间的包含关系。在此基础上，我们设计了一系列导航操作帮助用户基于层次概念格进行知识发现。为满足导航操作的实时性，我们提出了格挖掘算法LMiner。它以自顶向下和深度优先方式遍历生成格；通过倒排索引当前已生成的极小节点，进行高效的节点冗余检查和边的增量计算。实验结果表明，LMiner的速度远快于现有算法，而索引却小得多。 关键词：分面导航；层次概念格；频繁项集挖掘 Using Hierarchical Concept Lattice to Support Faceted Navigation HE Chao, CHENG Xue-Qi, GUO Jia-Feng Key Laboratory of Network Science and Technology, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190. E-mail: hechao@ Abstract: Faceted navigation is prevalent for searching and browsing multi-faceted resources。It recommends only the categories whose refining result of current search result is not empty, in order to help users understand what current search result is related to and avoid the dead end in the process of navigation. However, current approaches for faceted navigation can hardly expose deep relations between the recommended categories. In this paper, a hierarchical concept lattice is proposed which fully and concisely expresses the set containment relation between all kinds of search results. A series of navigation operations are proposed to help knowledge discovery based on the lattice. To guarantee the real-time property of these operations, a lattice mining algorithm LMiner is proposed. It generates all the nodes in a top-down and depth-first traversal of the whole lattice. By indexing all the minimal generated nodes in an inverted list, LMiner can accomplish subsumption check and incremental edge construction efficiently. Experimental results demonstrate that LMiner is much faster than current approaches while its index size is much less. Keywords: Faceted navigation; hierarchical concept lattice; frequent itemset mining 1 引 言 电子商务网站和数字图书馆通常采用多维分类树组织资源。对当前查询结果，用户选择新的类别对其细化或泛化。传统方式下，用