Periodical Intervals Mining Algorithm （PIM Algorithm）為一針對序列型樣於時間分佈用來挖掘潛在週期的演算法，藉由 PIM Algorithm 可針對時間間隔將週期分佈挖掘出來並分析出序列型樣的間隔天數。然而PIM Algorithm 並無針對單一消費行為趨勢作更詳細之探討，使得該週期分析所得結果之準確度會有較大誤差。此外，PIM Algorithm 僅能分析單純具有完全遞增/遞減型的購買週期分佈趨勢，然而在現實生活中，購買行為週期的趨勢分佈會呈現很大的變動性，具有完全遞增/遞減型的購買週期分佈只佔少數。

因此，為了提升週期分析的準確度與演算法的適用性，本論文修改原先PIM Algorithm對資料所作之預處理，針對產品重複購買行為趨勢重新定義，改良 PIM Algorithm 的缺點，提出一以『Divide and Conquer』為核心之完整週期分析演算法 Modified Periodical Intervals Mining Algorithm（MPIM Algorithm），分析消費者重複購買行為的週期，藉由所有產品之間序列的銷售週期比較出最佳推薦產品的銷售時間點以提供產品行銷的最有利資訊。

Periodical Intervals Mining Algorithm (PIM Algorithm) is an algorithm for analyzing the periodical properties of time intervals over sequential pattern mining. However, PIM Algorithm does not make a more detailed discussion on the purchase behavior, it will affect the accurate rate of periodicity detection. Moreover, PIM Algorithm is only suited for the pure ascending or descending type distribution function.
In a real-world scenario, purchasing behavior is extremely dynamic, it is only taken minority to fit the type distribution function mentioned above.

As a result, the aim of this work is to improve accuracy of the periodicity analysis and applicability of PIM algorithm. The study revises the preprocess of data, redefines the trend of the repeat buying behavior, and improves the shortcoming of PIM Algorithm. The Modified Intervals Mining Algorithm (MPIM Algorithm) takes the divide-and-conquer approach to collect the knowledge of the designated distribution function. A good agreement has been found between the analytical and experimental result shows good agreement.

目錄

第一章	緒論------------------------------------ 1
1-1 前言----------------------------------------	1
1-2 研究動機與目的-------------------------------- 3
1-3 研究架購-----------------------------------	6
第二章	文獻探討-------------------------------	7
2-1 序列型樣探勘-------------------------------	7
2-1-1 序列型樣法則探勘-------------------------	7
2-1-2 範例說明---------------------------------	12
2-2 Periodical Intervals Mining Algorithm------	20
2-2-1 LPDT與LPDM-------------------------------	21
2-2-2 PIM Algorithm----------------------------	27
2-2-3 PIM Algorithm範例說明--------------------	30
第三章	研究方法-------------------------------	32
3-1 問題定義-----------------------------------	32
3-2 MPIM Algorithm-----------------------------	36
第四章	實驗探討-------------------------------	42
4-1 實驗環境說明-------------------------------	42
4-2 重複性購買週期分析-------------------------	45
4-2-1 PIM Algorithm 與 MPIM Algorithm比較------	45
4-2-2 重複購買週期驗證-------------------------	47
第五章	結論與未來研究方向---------------------	49
參考文獻-------------------------------------	50
附錄-英文論文---------------------------------	52

圖目錄

Fig 1 <3304,3302> 透過PIM Algorithm分析報告----	5
Fig 2 <3304,3302> 透過MPIM Algorithm分析報告---	5
Fig 3 AprioriALL 演算法------------------------	9
Fig 4 將Lk-1合併得到Ck-------------------------	10
Fig 5 找出最大高頻序列-------------------------	11
Fig 6 LPDT虛擬碼-------------------------------	23
Fig 7 (24-x)/18(sin(x)+1.5) 曲線分佈圖---------	24
Fig 8 (24-x)/18(sin(x)+1.5) 透過LPDT求出之LIST- 25
Fig 9 LPDM虛擬碼-------------------------------	26
Fig 10 遞減性週期分佈曲線分佈圖----------------	27
Fig 11 PIM Algorithm---------------------------	29
Fig 12 < 3902,3103 >的趨勢分佈函式分佈圖-------	30
Fig 13 <3902,3103> 透過PIM Algorithm 分析報告--	31
Fig 14  2-序列分佈可能列表---------------------	33
Fig 15 MPIM Algorithm 虛擬碼-------------------	38
Fig 16 對2001年資料作AprioriALL algorithm 之結果 40
Fig 17 < 3103,2001 >趨勢分析圖型報告-----------	40
Fig 18 < 3304,2001 >於PIM Algorithm 分析結果---	46
Fig 19 < 3304,2001 >於MPIM Algorithm 分析結果--	46
Fig 20 <3302,3103>2001年分析報告---------------	47
Fig 21 <3302,3103>2001年分析報告---------------	47

表目錄

Table 1 交易紀錄-------------------------------	12
Table 2 產品交易統計---------------------------	13
Table 3作排序後之結果--------------------------	14
Table 4 顯著項目集-----------------------------	15
Table 5 轉換後的序列---------------------------	16
Table 6 產生之高頻序列組之一-------------------	17
Table 7 產生之高頻序列組之二-------------------	18
Table 8 MPIM Algorithm 初始值設定--------------	39
Table 9 實驗環境-------------------------------	43
Table 10 MPIM Algorithm 程式實作所需套件列表---	44

[1] R. Agrawal and T. Imielinski. (1993). Mining association rules be-tween sets of items in large databases. Presented at In Proceedings of ACM SIGMOD International Conference on Management of Data. 

[2] R. Agrawal and R. Srikant. (1995, Jan). Mining sequential patterns. Proceedings of the Eleventh International Conference on Data Engi-neering 

[3] J. Han, J. Pei, B. Mortazavi-Asl, Q. Chen and U. Dayal. (2000, Jan). FreeSpan: Frequent pattern-projected sequential pattern mining. Pro-ceedings of the Sixth ACM SIGKDD International Conference 

[4] M. Garofalakis, R. Rastogi and K. Shim. (2002). SPIRIT: Sequential pattern mining with regular expression constraints. Presented at Proceedings of the International Conference on very Large Data Bases. 

[5] M. Lin and S. Lee. (2002, Jan). Fast discovery of sequential patterns by memory indexing. Lecture Notes in Computer Science 

[6] J. Pei, J. Han, B. Mortazavi-Asl, H. Pinto and Q. Chen. (2001, Jan). PrefixSpan: Mining sequential patterns efficiently by prefix-projected pattern growth. Proceedings of the International Conference on Data Engineering 

[7] M. Zaki. (2001).SPADE: An efficient algorithm for mining frequent sequences. Mach. Learning 42(1), pp. 31-60. 

[8] J. Han, G. Dong and Y. Yin. (1999, Jan). Efficient mining of partial periodic patterns in time seriesdatabase. Data Engineering 

[9] J. Yang, W. Wang, P. Yu and J. Han. (2002). Mining long sequential patterns in a noisy environment. Proceedings of the 2002 ACM SIGMOD International Conference on Management of Data pp. 406-417. 

[10] O. Sridhar and R. Silberschatz. (1998, Jan). Cyclic association rules. Proc.14th International Conference on Data Engineering 

[11] D. Chiang, S. Lee, C. Chen and M. Wang. (2005). Mining interval sequential patterns. Int J Intell Syst 20(3), 

[12] Y. Chen. (2003, Oct). Discovering time-interval sequential patterns in sequence databases. Expert Syst. Appl. 25(3), pp. 343-354. 

[13] Y. Lee. (2005, Jul). 序列型樣之週期性間隔分析 The periodical intervals analysis on sequential patterns. pp. 1-98.