Interactive Association Rule Mining with Constraints

Title: Interactive Association Rule Mining with Constraints

Research Question: How can we efficiently support interactive mining sessions in the setting of association rule mining, while allowing users to specify conditions on the associations to be generated?

Methodology: The authors propose three approaches to support interactive mining sessions: integrated querying, post-processing, and incremental querying. In the integrated querying approach, constraints on the rules and sets to be generated are directly incorporated into the mining algorithm. In the post-processing approach, as much association as possible is mined first, and then standard lookups are performed on the set of materialized associations to answer user queries. The incremental querying approach combines the advantages of both previous approaches.

Results: The authors present an algorithm that significantly improves the efficiency of interactive mining sessions. They show that the querying achieved by exploiting constraints is optimal, meaning that it never generates an itemset that could give rise to a rule that does not satisfy the query. Furthermore, the number of generated itemsets during a query execution becomes proportional to the strength of the constraints in the query, making the execution faster as the query becomes more specific.

Implications: The presented algorithm not only reduces the number of passes through the database and the size of the database itself, but also reuses results from earlier queries to answer new queries, further improving efficiency. This approach provides a practical solution for supporting interactive data mining query languages, allowing users to more effectively explore and analyze their data.

Link to Article: https://arxiv.org/abs/0112011v2 Authors: arXiv ID: 0112011v2