Earticle

초록

영어

Concept hierarchies are important in many generalised data mining applications, such as multi-level association rule mining. There are drawbacks in using concept hierarchies constructed by domain experts. Therefore, we need automatic methods. We focus on situations where attributes of objects are useless for such a task. The attributes, such as price and size, are not relevant in the hierarchical structure of items when mining association rules from market basket data. Instead, the similarity between items from the point of view of customers is essential for organising the items. Assuming the availability of a distance matrix, our approach modifies the traditional hierarchical clustering algorithms to build a concept hierarchy. With some decision rules, a concept can be generalised from more than two specific concepts. Furthermore, the objects may be pre-processed, such that objects with high similarities are clustered into a ground concept. We formulate some evaluation criteria for the quality of constructed concept hierarchy.

목차

Abstract
 I. INTRODUCTION
 II. RELATED WORKS
 III. CONCEPT HIERARCHY CONSTRUCTION
 IV. MEASUREMENT METRICS
  A. Intrinsict Quality Measurement
  B. Indirect Quality Measurement
 V. PRE-PROCESSING
 VI. CONCLUSIONS
 REFERENCES
 BIOGRAPHIES

키워드

저자

• Huang-Cheng Kuo [ Department of Computer Science and Information Engineering, National Chiayi University, Chia-Yi City 600, Taiwan ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  JoC [Journal of Convergence]

• 간기

  계간

• pISSN

  2093-7741

• eISSN

  2093-775X

• 수록기간

  2010~2015

• 십진분류

  KDC 004 DDC 004