Information Technology Reference
In-Depth Information
In order to deal with thresholds on the rules relative to the actual values
reached in a rule set we introduce aggregate functions on rule quality measures.
The following query returns rules having rather low support and at the same
time quite high confidence compared to all other rules in the cache. min(supp)
denotes the minimum support value of all rules in the rule cache and max(conf)
the maximum confidence respectively.
SelectRulesFrom rulecache
supp < 1.1 * min(supp)and conf > 0.9 * max(conf;
Next, we want to specify rules based on certain items. For that purpose we
introduce the keywords assumption and consequent to address the items in the
body and respectively the head of the rules. The following query retrieves rules
containing a certain item in the rule consequent.
SelectRulesFrom rulecache
Where '2ndAirbag' in consequent and conf > 0.85;
While we do not see the need for explicit
∃
and
∀
quantifiers on sets of
rules, having these on the itemsets is quite useful. In conjunction with the at-
tributes linked to items, powerful queries become possible. For example we could
select all rules that have support higher than 5%, confidence higher than 95%,
and that “explain” at least one special equipment which incurs costs above a
certain threshold. In the following x.attributename gives the value of attribute
attributename for item
x
:
SelectRulesFrom rulecache
Where supp > 0.05 and conf > 0.95
and exists x in consequence (x.type = 'spec equip'
and x.costs > 1000);
A more complex and also very useful query is to find all rules with nothing but
special equipment in the consequence originating from a manufacturer who also
manufactures at least one special equipment from the rule assumption.
SelectRulesFrom rulecache
Where forall x in consequence (x.type = 'spec equip'
and exists y in assumption (y.type = 'spec equip' and
x.manu = y.manu));
The quantifiers and attributes on the items and the aggregate functions on
the rules are both intuitive to use and flexible. The examples above give a first
impression of their potentials.
7 Summary
In this paper we dealt with association rule mining in the context of a complex,
interactive and iterative knowledge discovery process. We introduced the basics
of association rules and of the algorithmic aspects of association rule mining.
