The Design and Application of an Automatic Link Analysis Technology

• Main Authors: Jing-Rong Yang, 楊境榮
• Other Authors: Hown-Wen Chen
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/40551157509865502945

碩士 === 大葉大學 === 資訊管理學系碩士班 === 93 === Each data mining technique is aimed to automatically analyze implied knowledge rules from data. However, the less frequent but high-value data association rules can not be extracted by simply setting a single proper threshold in algorithms of association rule analysis. For solving the kind of problems, data must be classified and then processed with different lowest support values. Unfortunately, the performance of handling many actual problems is still unacceptable. Besides, human experts need to use visualization tools to find out regular patterns and features with their eyes in traditional link analysis. An only exceptional case is Google, which takes Page Rank algorithm to automatically evaluate the weight of each network by hyperlink relations. In other words, an automatic technique for link analysis is eagerly needed for problems such as food chain, transportation network, etc. Some researches of social network analysis pointed out “strong ties within a graph can group individuals of the same characteristic while weak ties can communicate and work as the bridge of different groups.” Based on the concept of weak ties and group theory, we proposed to find out potential weak links beyond biconnected and strongly connected components and then form critical paths within a graph . The proposed algorithm can detect out association relations between rare critical data which is quite difficult to deal with in traditional association rule analysis. In order to verify and evaluate proposed automatic link analysis, the actual Enron Email Dataset announced by FERC(Federal Regulation and Oversight of Energy) was investigated. Experiments illustrated the efficiency of the algorithm in analyzing characteristics of a direct/undirected graph. Thus, it is highly recommended to solve problems such as detection of social group, organizational criminality, e-mail spam etc.