Hierarchical Online Mining for Associative Rules

Mining for associative rules in transaction databases has been a very active area of research in recent years; purchases made by customers at a store provide one example of such transactions. The typical multi-pass associative rule mining algorithm finds large itemsets in the first phase, using a specified minimum support level, and then discovers associative rules in the second phase. In this paper, we propose and discuss an online algorithm which is based on hierarchical classification of items. The proposed algorithm carries out the first phase efficiently in one pass, with tight bounds on the computational effort required, and modest memory requirements. The algorithm is thus capable of online mining of associative rules from transaction streams. We also discuss in the paper some practical issues related to applications of associative rule mining, and the consequent need to broaden the definition of an associative rule.