[[alternative]]An Empirical Study for Mining Association Rules on Individal Cluster

碩士資訊工程學系[[abstract]]分群法(clustering)及關聯性法則探勘(association rules mining)是資料探勘領域中兩種重要的方法。分群法主要用來分析尚未整理以及難以看出規則的原始資料，將較相近的資料分為同一群，並藉由分群結果可以較容易的分析其內容，然而當資料越複雜時，分析每一群組特徵的困難度就越高。而關聯性法則是探索物件屬性項目同時出現的情形，來發覺屬性間的關聯性，在這過程中若資料量太大則所產生的法則數量可能會太多，或是出現頻率太少的屬性可能會被忽略掉，這造成關聯性法則探勘的一大缺失。 104家教網是國內著名找老師找家教的網站，在本研究中將利用分群法及關聯性法則探勘二項方法，來探討使用者的使用行為。在此研究先使用分群法將較接近的資料分為同一群，再分別對已經減少了資料量的每一群作關聯性法則探勘。研究發現此方法結合了兩種方法的優點，改進了僅使用單一方法的缺點，能讓使用者得到更多的資訊，來更容易的分析使用者的使用行為。[[abstract]]Clustering and association rules mining are two important methods in data mining. Clustering is used to divide the similar data into the same group. By analyzing these formed groups, users can easily abtain the hidden pattern from data. However when datasets are large and complicated, analyzing characteristic of each group will become difficult. Association rules mining can find out frequent itemsets that items always appear in the same transaction. The results of mined rules will show the relationships among items. Thus, there are some weaknesses for association rules mining. When datasets are large, it maybe produces a large amount of rules. The other weakness is that items with few appearance in transactions will be eliminated during finding frequent items process; therefore information about these items will be unavaiable. In this paper, we apply clustering and association rules mining on data collected from http://ww... to analyaze users’ behaviors. First, using clustering algorithm divide the dataset into several groups. Then using association rules mining find out important rules from each groups. The experimental evidence shows that combining these two approachs can improve the disvantages of adopting clustering or association rules mining alone. Therefore it will help users to analyze data easily and efficiently