During any association mining process it is a big challenge to remove uninteresting rules. We are interested in effective formal and experimental method for finding interestingness of the multilevel rules.
.
this might be of interest to you :
A Bayesian approach for classification rule mining in quantitative databases
Dominique Gay, Marc Boullé
Machine Learning and Knowledge Discovery in Databases
Springer, 2012
http://scholar.google.fr/citations?view_op=view_citation&hl=fr&user=IdzmYQcAAAAJ&citation_for_view=IdzmYQcAAAAJ:eQOLeE2rZwMC
.
If you can add semantic labels to your items, you can often use those to filter out uninteresting rules by 1) deleting rules with items of no interest, or 2) use the labels to design a ranking function. The ranking function can be topic specific (if you want a certain domain/topic) or generic (e.g., if you want general or specific items).
You could also try using confidence and lift values for association rules.
lets say we have a rule: X -> Y.
The confidence is the ratio of the transactions that contain XUY to the total transactions that contain X.
The lift value of a rule tells the degree to which Y is more likely to co-occur when X occurs and can be used as strength of a rule. A lift value greater than 1.0 implies positive association.
You could read this paper:
http://www.oalib.com/paper/2651823#.VEkLAiKsXKM
May be it helps you
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining