A Study of Mining Sequential Pattern in Large Dynamic Transaction Database

• Main Authors: Shen-jung Hsu, 徐勝榮
• Other Authors: Chien-I Lee
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/44019394452474051290

碩士 === 國立臺南大學 === 資訊教育研究所碩士班 === 92 === Mining sequential pattern plays an important role in data mining. Numerous algorithms have been proposed to mining sequential patterns efficiently in a static database. However, the maintenance of such discovered sequential patterns is nontrivial in large database. In real world, a transaction database may allow the users to insert/delete the transaction data and the customer data frequently. Additionally, the development of multiple shop leads in the urgent requirement of fast mining sequential patterns in the merged database. Therefore, the efficient maintenance of the latest sequential patterns in dynamic large database becomes more and more important. First, this study presents the incremental sequential patterns mining (ISP) algorithm to maintain the discovered sequential patterns in incremental transaction database efficiently. ISP applies the information of the discovered sequential patterns in original database to reduce the number of candidates. Consequently, the performance of ISP is better than that of GSP in our experimental results. Furthermore, this study also proposes the updated sequential patterns mining (USP) algorithm, which extends the concept of reducing the number of candidates to support both insertion and deletion operations in an update database efficiently. For the issue of efficient mining sequential patterns in the merged database, this study develops the merged sequential patterns mining (MSP) algorithm. Contract to re-mine a merged large database, MSP uses two small database, which have discovered their sequential pattern, respectively to speed up the mining process. Simulation results reveal that ISP, USP and MSP are superior to GSP, respectively in several artificial datasets. The more difference of the frequent sequential pattern between original database and incremental database, the more performance difference between ISP and GSP. USP is very efficient when the proportion of the updated part of the database is small. The performance difference is greater between MSP and GSP when the proportion of the common frequent 1-items in two unmerged database is smaller. Simulation results also appear that the three algorithms use less memory than that of GSP, respectively.