自動鏈結分析演算法在社會網絡之開發與應用

在關聯分析或序列樣式分析之資料探勘研究中，即使採用了多重門檻值的設定來過濾大資料集合，仍會找到過多無用且信度過低的關聯法則，或可能遺漏了頻率較低但實質上卻具有高度價值的資料項目。此外，除了少數特定問題外，以往鏈結分析之研究，都需要仰賴專家來目測已轉換為視覺化的資料，來進行主觀評估，以發現資料之規則性。此種衡量和評估的方式，對於複雜之網絡，往往費事耗時且成效不彰。而一些社會網絡之研究也指出，頻率低的弱鏈結扮演著聯繫不同群體之重要角色。因此，本研究透過基本的圖學理論，提出一個不需要依賴門檻值設定，就能找出存在於網絡中的弱鏈結及關鍵弱鏈結路徑之自動鏈結演算法。本研究再利用真實的安隆企業電子郵件資料，配合NetDraw視覺化的網絡分析工具，以實驗來檢驗本自動化鏈結分析演算法之可行性及正確性。

關鍵字

資料探勘；鏈結分析；弱鏈結；關聯法則

並列摘要

Even with the settings of multiple thresholds when screening large data sets, using link analysis or sequential patterns analysis, many data mining studies obtain lots of not-very-useful low-confidence association rules or miss the low-frequency but actually highly valuable data items. In addition, except for some specific problems, previous link analysis researches mostly rely on experts´subjective visual investigations of analyzed data which are transformed into visual form in order to find the data's regularities. This kind of assessment and evaluation is usually time-consuming and inefficient for complicated networks. Prior studies of social networks have revealed that low-frequency (weak) links play important roles in connecting different cliques in a social network. Therefore, utilizing the topology in graph theory this study proposes an automatic link analysis algorithm without depending on the thresholds to discover the weak links and the key weak link paths in a network. To check and see the feasibility and accuracy of the proposed algorithm, empirical studies on the well-known Enron e-mail data sets using the NetDraw network visualization tool are conducted, and the results are found to be positive.