Association Rule Mining with Constraints Rebuttal Deepen Manek

Association Rule Mining with Constraints - Rebuttal Deepen Manek Kirsten Rutherford October 27, 2003

Association Rule Mining with Constraints • What about constraints with taxonomies? Also, are referential integrity type of constraints handled (e. g. manager and employee in same itemset)? • On p. 6, is the example really an association rule? • On p. 8, how is mining frequent itemsets different than mining association rules? • On p. 11, aren’t we dealing with mining association rules with constraints, not simply mining frequent itemsets? • On p. 12, how are convertible constraints used in determining the association rules? • On p. 12, is an inconvertible constraint a ‘catch-all’ for all constraints that don’t fall into other categories?

Association Rule Mining with Constraints • On p. 13, are we dealing with subsets or supersets? Also, on p. 14, is Min(S) >= 7 the constraint? • On p. 15, can you define selection predicate, powerset, and SATC ? Also , on p. 18, can you please clarify how C 1 is succinct? • On p. 22, what does weakly mean? What about equality constraints (e. g min(S)=v)? Also, can you explain what range(S) means? • On p. 24, can you explain how avg(S) is both convertible monotone and convertible anti-monotone? Also, how do we define monotone constraints? • On p. 38, are we supposed to find the frequent items of the set {1, 2, 3, 4}? Also, is this Strategy III as mentioned on p. 34?

Association Rule Mining with Constraints • On p. 43, we are only looking at a succinct, anti-monotone constraint? What is the performance of CAP for constraints that aren’t succinct? • On p. 43, has any performance measurements been done with the FP-growth algorithm? Also on p. 43, we are dealing with a 1 -var constraint. What is performance like if there are several constraints? • On p. 45, are you saying that CAP generates less frequent item-sets than it should? • Is it appropriate to push constraints into the association rule algorithms? If we had the processing time, shouldn’t we generate all the associate rules for future reference? • Do you think it would be useful to split the mining process into phases?