Building Concept Hierarchy with Combination of Multiple Members

• Main Authors: Yen-Chang Su, 蘇彥彰
• Other Authors: Huang-Cheng Kuo
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/77763944119200120562

碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 102 === In data mining research area, concept hierarchies are so popular theme. We get distance between object and the other object, and the smaller the distance is, the more similar the concept is. Next we put similar concepts into a category, and put similar categories into a super category. Further, the process is recursively conducted until there is only one category left, which is the root of the hierarchy. Even it is popular, there are many challenges in constructing concept hierarchy. Therefore, we propose a new method in building the concept hierarchy. General method is considering merger of two clusters by hierarchical agglomerative clustering algorithm. However, there are a few practical examples, such as buying flashlights, batteries, and electric razors, but these three items may not be placed in the same location. We try to design different merging method and compare the other methods. If the structure can merge on some situations, we can use measurement metrics which propose in past. We use the threshold value to control the structure of concept hierarchies and present our results to compare the quality among different methods. Finally, our purposed method is modified from hierarchy agglomerative clustering (HAC), and is different to past by combining iterations of merging/joining more than or equal to two clusters.