Association Rule Mining

Title: Association Rule Mining

Main Research Question: How can we support interactive mining sessions in the setting of association rule mining?

Methodology: The authors propose a combination of incorporating filtering conditions within the mining phase and filtering already generated associations. They present several concrete algorithms and compare their performance.

Results: The authors find that the filtering achieved by exploiting the query conditions is non-redundant, meaning it never generates an itemset that, apart from the minimal support and confidence thresholds, could result in a rule that does not satisfy the query. They also note that the number of passes through the database and its size are reduced, proportionally to the strength of the conditions in the query. Furthermore, they state that a generated itemset will never be regenerated as a candidate itemset.

Implications: The authors' approach allows for more efficient interactive mining sessions in the setting of association rule mining. This can lead to faster execution of queries and reduced database usage. Additionally, their method ensures that the results of earlier queries are not redundantly used in the generation of new itemsets.

Link to Article: https://arxiv.org/abs/0112011v1 Authors: arXiv ID: 0112011v1