| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 90 === Sequential patterns are useful for analyzing the purchasing behaviors of Customers. Previous mining algorithms on sequential patterns consider only the purchasing order between two itemsets. In some applications, sequential patterns with quantities may be useful. A sequential pattern takes the form of {X}à{Y}, which means that if a customer buys itemset X then, most of the time, he will also buy itemset Y later. We define a quantitative sequential pattern as a sequential pattern in which both the antecedent and the consequence are associated with a quantity interval. One example of quantitative sequential pattern is B[10, 12]àA[17,18], which means that if a customer buys 10 to 12 units of B, he will also buy 17 to 18 units of A later.
In this Thesis, we develop algorithms for mining quantitative sequential patterns. To find quantitative sequential patterns, we first apply the Boolean algorithm to derive the set of all sequential patterns. Then for a selected sequential pattern, we use partition to derive the associated quantity intervals. Two partition methods have been proposed. One of them is fuzzy partition while the other is direct partition. Experiments show that both methods find all valid quantity intervals. Besides, the direct partition is more efficient while the fuzzy partition is less sensitive to the initial partition unit.
|
|---|
