Using Data Mining Techniques to Identify Substitute and Complementary Relationships among Products

• Main Authors: Chu-hui Liao, 廖珠惠
• Other Authors: Yung-Ho Leu
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/59439896429053313553

碩士 === 國立臺灣科技大學 === 資訊管理系 === 90 === The complementary relationship and the substitute relationship among product items play important roles in marketing. In this thesis, we try to find the complementary relationship and the substitute relationship among product items by analyzing the users’ purchasing behaviors in the transactional database. We modify the Apriori algorithm to find the complementary relationship. The modification is along two directions. First, we differentiate items between durable items and non-durable items. Second, we consider not only items that are bought in the same transaction but also items that are bought by the same customer in different transactions. The rationale behind the modification is as follows. First, a durable item often occurs infrequently; therefore, it cannot be included in the rule generation in the Apriori algorithm. However, the complementary relationship between durable items and non-durable items does exist. Second, the Apriori algorithm only considers items that are bought in the same transaction. While analyzing complementary relationship, one should also consider items that are bought by the same customer in different transactions. The complementary relationship mined by the modified Apriori algorithm agrees better with the definition of complementary product items in economics. The mining algorithm for substitute items is based on the concept hierarchy of product items. According to the concept hierarchy, three different types of substitute relationship can be found. They are substitute relationship between brands, substitute between products and substitute between categories. We have implemented the proposed algorithms and test them against the FoodMart database, a sample database in SQL Server. The experiment results show that our algorithm can effectively find the complementary relationship and the substitute relationship among product items. Keywords : Data Mining, Concept Hierarchy, Complementary Relationship, Substitute Relationship, Attractive Itemsets