610 likes | 2.22k Views
15/Sep/2006 04:00 PM. S.P.Vimal, CS IS Group, BITS-Pilani. 2. To discuss
E N D
1. Association Rule MiningConstraint Based Association Rule Mining
S.P.Vimal
Assistant Lecturer
CSIS/BITS-Pilani
vimalsp@bits-pilani.ac.in
2. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 2 To discuss…
Constraint Based Association Rule Mining
Concepts
Metarule-Guided Rule mining
Constraint pushing
Types of rule constraints
antimonotonic
monotonic
succinct
convertible
inconvertible
3. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 3 Constraint Based Rule Mining - Concepts
Constraints ?
Users expectation or intuition helps confine the search space
Forms of constraints
Knowledge type constraints
Data constraints
Dimension / Level constraints
Interestingness constraints
Rule constraints
4. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 4 Constraint Based Rule Mining - Concepts
Rule constraints ?
Specify the ‘form of the rules’
Rules take the form
Rule Template / Meta Rule
Set/subset relationships of attributes mined, aggregates etc.
‘Mining query optimizer’ must be incorporated in the mining process to exploit the constraints specified
5. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 5 Constraint Based Rule Mining - Metarule-Guided
Specifies the syntactic form of the rules, interested
Syntactic forms serves as the constraint
Based on analysts experience, expectation, or intuition regarding data
To analyze the customers traits leading to the purchase of office software, meta rule will be
P1(X,Y) ? P2(X,Z) ? buys (X, ”office software”)
where P1,P2 are the predicates on customer X
6. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 6 Constraint Based Rule Mining - Metarule-Guided
Data mining system searches for the rule of the form that matches the meta rule given
For ex. The rule generated matching the given metarule is
age (X, “30..40”) ? income (X, “30K..50K”)
? buys (X, “office software”)
7. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 7 Constraint Based Rule Mining - Metarule-Guided
Consider the template
P1 ? P2 ? … ? Pl ? Q1 ? Q2 ?… ? Qr
Each Pi’s and Qj’s are predicates (instantiated / variables)
and l + r = p
To mine for the rules satisfying this template
Find all frequent p-predicate sets, Lp
Find support & confidence of Lp
8. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 8 Constraint Based Rule Mining - Constraint Pushing
Allows pushing constraints deep into mining process to confine the search space, assuring the completeness of the result as well
Rule constraints specified as expected set/subset relationship of the variables involved, aggregate functions etc
Can be used in conjunction with metarule-guided mining
9. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 9 Constraint Based Rule Mining - Constraint Pushing
Look at the following scenario
A datawarehouse with
Fact table : sales (cust_name, item_name, TID)
Dimension Tables :
lives_in (cust_name, region, city)
item (item_name, region, city)
transaction (TID, day, month, year)
And the mining query
“Find the sales of which cheap items (price<100$) promote sales of expensive items (price>500$) of the same group for delhi customers in 2004”
10. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 10 Constraint Based Rule Mining - Constraint Pushing
The DMQL query above case would be
mine association as
lives_in(C,_, “delhi”) ? sales+ (C, ?{ I}, {S}) ? sales+ (C, ?{ I}, {S} )
from sales
where S.year=2004 and T.year=2004 and I.group=J.group
group by C, I.group
having sum(I.price) < 100 and min (J.price)>500
with support threshold = 1%
with confidence threshold = 50%
11. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 11 Constraint Based Rule Mining - Constraint Pushing
From this DMQL Query we can deduce the following constraints specified
Meta Rule : Line 2
Knowledge constraint : Line 1
Data constraint : Line 3, line2
Level constraint : Line 2
Rule constraint : Line 4 and Line6
Interestingness
Constraint : Line 8
12. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 12 Constraint Based Rule Mining - Types of Rule Constraints
Rule constraints can be categorized as
antimonotonic
monotonic
succinct
convertible
inconvertible
Ensures completeness of result while pushing these rules deep into the mining process
13. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 13 Constraint Based Rule Mining - Types of Rule Constraints
antimonotonic
“if a itemset does not satisfy the rule constraint, then none of its supersets satisfy” , property of antimonotonic rules
example : sum ( I.price >100)
count ( I ) < 100
avg ( I ) < 250 is not antimonotonic
Note, apriori property is antimonotonic.
14. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 14 Constraint Based Rule Mining - Types of Rule Constraints
monotonic
“if a itemset satisfy the rule constraint, then all of its supersets satisfy” , property of monotonic rules
Example : sum (I.price) > 100
v ? S
min(S) = V is not monotonic
Once the subset satisfies this property, further testing for this rule is redundant
15. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 15 Constraint Based Rule Mining - Types of Rule Constraints
succinct
“All and only those set guaranteed to satisfy the rule can be enumerated” Property of succinct rules
The itemsets can be generated that satisfy the rule even before the support count starts
Once such subset is generated, iterative testing for the constraint can be effectively avoided
Example : min(J.price) > 500
max(S) < 120
avg(S) > v , avg(S) <v are not succinct
16. 15/Sep/2006 04:00 PM S.P.Vimal, CS IS Group, BITS-Pilani 16 Constraint Based Rule Mining - Types of Rule Constraints
convertible constraints
Constraints not satisfying to any of antimonotonic, monotonic, succinct can be made to satisfy antimonotonic, monotonic constraints by changing order of elements in the set
Ex : Avg(price) < 100
Inconvertible
Constraints which are not convertible
Ex : Sum(S) < v , sum (S) > V ,
element of set S could be any real value