Finding Cross-Object Relationships from Large Databases

While traditional association rules demonstrate strong potential values such as to improve market strategies for retail industry, they are limited to finding associations among items within the same transaction. Consider a database of supermarket transactions, the traditional association rules can represent such knowledge as “80% of customers who buy Chinese tea also buy teapot at the same time.” However, they fail to represent some more interesting rules like “If a customer buys Chinese tea, s/he may most likely buy teapot within 3 days”, where the association may span across different transactions. To capture this contextual semantics which are also vital to the validation of associations, in this study, we introduce the notion of cross-object relationships. Two algorithms for mining cross-object association rules from large databases are developed by extension of Apriori algorithm. We show that traditional associations can be treated as a special case of cross-object relationships from both conceptual and algorithmic point of view.