Mining transaction data by extracting rules to express relationships between itemsets is a classical form of data mining. The rule evaluation method used dictates the nature and the strength of the relationship, eg. an association, a correlation, a dependency, etc. The widely used Apriori algorithm employs breadth-first search to find frequent and confident association rules. The Multi-Stream Dependency Detection (MSDD) algorithm uses iterative deepening (ID) to discover dependency structures. The search bound for ID can be based on various characteristics of the search space, such as a change in the tree depth (MSDD), or a change in the quality of explored states. This paper proposes an ID-based algorithm, ID-based algorithm, ID_(G_(max)), whose search bound is based on a desired quality of the discovered rules. The paper also compares strategies to relax the search bound and shows that the choice of this relaxation strategy can significantly speed up the search which can explore all possible rules.
展开▼
