在資料探勘領域中,關聯規則大多用來分析交易紀錄中顧客購買商品的行為,而網路使用行為探勘(web usage mining)則用以分析使用者瀏覽網頁的瀏覽行為,在數位學習管理系統上也蘊藏著大量的學習者行為,因此本研究利用關聯規則來探勘e-Learning資料庫以擷取出學習者的行為樣式。在關聯規則探勘中最常見且普遍的是使用Apriori演算法,有學者提出利用此演算法結合多支持度門檻值、階層類別關聯規則及量化關聯規則來挖掘交易資料庫,但Apriori演算法在低門值、長字元(long pattern)及大量的頻繁項目(frequent pattern)狀態下是相當沒有效率地。因此本研究主要應用P-tree與FP-tree相似結構提出一個新的MFMFP-tree結構及MFMQFP-Growth演算法,來從多支持度量化資料庫中挖掘出多支持度模糊關聯規則,並應用在探勘學習者的學習行為,透過數位學習管理系統的紀錄檔中,擷取出學習者選擇教材時的學習路徑,以提供下一個學習者選課上的建議,最後提出在新的演算法下支持度門檻值的調整機制。
In the field of data mining, association rules are used to analyze customers’ relations of the transaction database, and web usage mining is to visit the website of the user browsing logs. In the learning management system also contains a large number of learners browsing logs, so the use of association rules on e-Learning to explore the database to retrieve a learner's behavior patterns. The association rules mining in the most common and widespread is the use of Apriori algorithms, some related researches applied about mining transaction database on multiple-level and quantitative association rules with multiple minimum supports. But their proposed algorithm based on Apriori algorithm that is an un-efficient on mining lower support threshold, long patterns and huge number of frequent patterns. So this thesis was to use P-tree and FP-tree like structure to propose a new MFMFP-tree and MFMQFP-Growth algorithms to mining multi-level fuzzy association rules from quantitative transactions with multiple minimum supports, and application on mining frequent patterns of learners’ behavior. Through learning management system to retrieve learning path on learner choose materials, to provide recommendation for next learning course. Finally, we also proposed support tuning mechanism under the new algorithms.