Data Mining

Session 6 – Main Theme

Mining Frequent Patterns,
Association, and Correlations

Dr. Jean-Claude Franchitti

New York University

Computer Science Department
Courant Institute of Mathematical Sciences
Adapted from course textbook resources
Data Mining Concepts and Techniques (2nd Edition)
Jiawei Han and Micheline Kamber

Agenda

11 Session
Session Overview
Overview

Mining
Mining Frequent
Frequent Patterns,
Patterns,
22 Association,
Association, and
and Correlations
Correlations

33 Summary
Summary and
and Conclusion
Conclusion

2
What is the class about?

ƒ Course description and syllabus:

» http://www.nyu.edu/classes/jcf/g22.3033-002/
» http://www.cs.nyu.edu/courses/spring10/G22.3033-002/index.html

ƒ Textbooks:
» Data Mining: Concepts and Techniques (2nd Edition)
Jiawei Han, Micheline Kamber
Morgan Kaufmann
ISBN-10: 1-55860-901-6, ISBN-13: 978-1-55860-901-3, (2006)
» Microsoft SQL Server 2008 Analysis Services Step by Step
Scott Cameron
Microsoft Press
ISBN-10: 0-73562-620-0, ISBN-13: 978-0-73562-620-31 1st Edition (04/15/09)

Session Agenda

ƒ Basic concepts and a roadmap

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

4
Icons / Metaphors

Information

Common Realization

Knowledge/Competency Pattern

Governance

Alignment

Solution Approach

55

Agenda

11 Session
Session Overview
Overview

Mining
Mining Frequent
Frequent Patterns,
Patterns,
22 Association,
Association, and
and Correlations
Correlations

33 Summary
Summary and
and Conclusion
Conclusion

6
Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

What Is Frequent Pattern Analysis?

ƒ Frequent pattern: a pattern (a set of items, subsequences,

substructures, etc.) that occurs frequently in a data set
ƒ First proposed by Agrawal, Imielinski, and Swami [AIS93] in
the context of frequent itemsets and association rule mining
ƒ Motivation: Finding inherent regularities in data
ƒ What products were often purchased together?— Beer and diapers?!
ƒ What are the subsequent purchases after buying a PC?
ƒ What kinds of DNA are sensitive to this new drug?
ƒ Can we automatically classify web documents?

ƒ Applications
ƒ Basket data analysis, cross-marketing, catalog design, sale campaign
analysis, Web log (click stream) analysis, and DNA sequence
analysis. 8
Why Is Freq. Pattern Mining Important?

ƒ Freq. pattern: An intrinsic and important property

of datasets
ƒ Foundation for many essential data mining tasks
ƒ Association, correlation, and causality analysis
ƒ Sequential, structural (e.g., sub-graph) patterns
ƒ Pattern analysis in spatiotemporal, multimedia, time-
series, and stream data
ƒ Classification: discriminative, frequent pattern analysis
ƒ Cluster analysis: frequent pattern-based clustering
ƒ Data warehousing: iceberg cube and cube-gradient
ƒ Semantic data compression: fascicles
ƒ Broad applications
9

Basic Concepts: Frequent Patterns

Tid Items bought ƒ itemset: A set of one or more

10 Beer, Nuts, Diaper items
20 Beer, Coffee, Diaper ƒ k-itemset X = {x1, …, xk}
30 Beer, Diaper, Eggs
ƒ (absolute) support, or, support
40 Nuts, Eggs, Milk count of X: Frequency or
50 Nuts, Coffee, Diaper, Eggs, Milk occurrence of an itemset X
Customer Customer
ƒ (relative) support, s, is the
buys both buys diaper fraction of transactions that
contains X (i.e., the probability
that a transaction contains X)
ƒ An itemset X is frequent if X’s
support is no less than a minsup
Customer
buys beer
threshold

10
Basic Concepts: Association Rules

Tid Items bought ƒ Find all the rules X Æ Y with

10 Beer, Nuts, Diaper
minimum support and confidence
20 Beer, Coffee, Diaper
30 Beer, Diaper, Eggs ƒ support, s, probability that a
40 Nuts, Eggs, Milk transaction contains X ∪ Y
50 Nuts, Coffee, Diaper, Eggs, Milk ƒ confidence, c, conditional probability
Customer
that a transaction having X also
Customer
buys both
buys
contains Y
diaper ƒ Let minsup = 50%, minconf = 50%
ƒ Freq. Pat.: Beer:3, Nuts:3, Diaper:4,
Eggs:3, {Beer, Diaper}:3
Customer
buys beer
ƒ Association rules: (many more!)
ƒ Beer Æ Diaper (60%, 100%)
ƒ Diaper Æ Beer (60%, 75%)

11

Closed Patterns and Max-Patterns

ƒ A long pattern contains a combinatorial number of sub-

patterns, e.g., {a1, …, a100} contains (1001) + (1002) + … +
(110000) = 2100 – 1 = 1.27*1030 sub-patterns!
ƒ Solution: Mine closed patterns and max-patterns instead
ƒ An itemset X is closed if X is frequent and there exists no
super-pattern Y ‫ כ‬X, with the same support as X
(proposed by Pasquier, et al. @ ICDT’99)
ƒ An itemset X is a max-pattern if X is frequent and there
exists no frequent super-pattern Y ‫ כ‬X (proposed by
Bayardo @ SIGMOD’98)
ƒ Closed pattern is a lossless compression of freq. patterns
ƒ Reducing the # of patterns and rules

12
Closed Patterns and Max-Patterns

ƒ Exercise. DB = {<a1, …, a100>, < a1, …, a50>}

ƒ Min_sup = 1.
ƒ What is the set of closed itemset?
ƒ <a1, …, a100>: 1
ƒ < a1, …, a50>: 2
ƒ What is the set of max-pattern?
ƒ <a1, …, a100>: 1
ƒ What is the set of all patterns?
ƒ !!

13

Computational Complexity of Frequent Itemset Mining

ƒ How many itemsets are potentially to be generated in the worst

case?
ƒ The number of frequent itemsets to be generated is senstive to the
minsup threshold
ƒ When minsup is low, there exist potentially an exponential number of
frequent itemsets
ƒ The worst case: MN where M: # distinct items, and N: max length of
transactions
ƒ The worst case complexty vs. the expected probability
ƒ Ex. Suppose Walmart has 104 kinds of products
ƒ The chance to pick up one product 10-4
ƒ The chance to pick up a particular set of 10 products: ~10-40
ƒ What is the chance this particular set of 10 products to be frequent
103 times in 109 transactions?

14
Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

15

The Downward Closure Property and Scalable Mining Methods

ƒ The downward closure property of frequent

patterns
ƒ Any subset of a frequent itemset must be frequent
ƒ If {beer, diaper, nuts} is frequent, so is {beer, diaper}
ƒ i.e., every transaction having {beer, diaper, nuts} also
contains {beer, diaper}
ƒ Scalable mining methods: Three major approaches
ƒ Apriori (Agrawal & Srikant@VLDB’94)
ƒ Freq. pattern growth (FPgrowth—Han, Pei & Yin
@SIGMOD’00)
ƒ Vertical data format approach (Charm—Zaki & Hsiao
@SDM’02)

16
Apriori: A Candidate Generation & Test Approach

ƒ Apriori pruning principle: If there is any itemset

which is infrequent, its superset should not be
generated/tested! (Agrawal & Srikant @VLDB’94,
Mannila, et al. @ KDD’ 94)
ƒ Method:
ƒ Initially, scan DB once to get frequent 1-itemset
ƒ Generate length (k+1) candidate itemsets from length k
frequent itemsets
ƒ Test the candidates against DB
ƒ Terminate when no frequent or candidate set can be
generated
17

The Apriori Algorithm—An Example

Supmin = 2 Itemset sup

Itemset sup
Database TDB {A} 2 L1 {A} 2
Tid Items C1 {B} 3
{B} 3
10 A, C, D {C} 3
20 B, C, E
1st scan {C} 3
{D} 1
{E} 3
30 A, B, C, E {E} 3
40 B, E
C2 Itemset sup C2 Itemset
L2 {A, B} 1
Itemset sup 2nd scan {A, B}
{A, C} 2
{A, C} 2 {A, C}
{A, E} 1
{B, C} 2
{B, C} 2 {A, E}
{B, E} 3
{B, E} 3 {B, C}
{C, E} 2
{C, E} 2 {B, E}
{C, E}
C3 Itemset 3rd scan L3 Itemset sup
{B, C, E} {B, C, E} 2

18
The Apriori Algorithm (Pseudo-Code)

Ck: Candidate itemset of size k

Lk : frequent itemset of size k

L1 = {frequent items};
for (k = 1; Lk !=∅; k++) do begin
Ck+1 = candidates generated from Lk;
for each transaction t in database do
increment the count of all candidates in Ck+1 that are
contained in t
Lk+1 = candidates in Ck+1 with min_support
end
return ∪k Lk;
19

Implementation of Apriori

ƒ How to generate candidates?

ƒ Step 1: self-joining Lk
ƒ Step 2: pruning
ƒ Example of Candidate-generation
ƒ L3={abc, abd, acd, ace, bcd}
ƒ Self-joining: L3*L3
ƒ abcd from abc and abd
ƒ acde from acd and ace
ƒ Pruning:
ƒ acde is removed because ade is not in L3
ƒ C4 = {abcd}

20
How to Count Supports of Candidates?

ƒ Why counting supports of candidates a

problem?
ƒ The total number of candidates can be very huge
ƒ One transaction may contain many candidates
ƒ Method:
ƒ Candidate itemsets are stored in a hash-tree
ƒ Leaf node of hash-tree contains a list of itemsets and
counts
ƒ Interior node contains a hash table
ƒ Subset function: finds all the candidates contained in
a transaction

21

Example: Counting Supports of Candidates

Subset function
Transaction: 1 2 3 5 6
3,6,9
1,4,7
2,5,8

1+2356

13+56 234
567
145 345 356 367
136 368
357
12+356
689
124
457 125 159
458

22
Candidate Generation: An SQL Implementation

ƒ SQL Implementation of candidate generation

ƒ Suppose the items in Lk-1 are listed in an order
ƒ Step 1: self-joining Lk-1
ƒ insert into Ck
ƒ select p.item1, p.item2, …, p.itemk-1, q.itemk-1
ƒ from Lk-1 p, Lk-1 q
ƒ where p.item1=q.item1, …, p.itemk-2=q.itemk-2, p.itemk-1 < q.itemk-1
ƒ Step 2: pruning
ƒ forall itemsets c in Ck do
ƒ forall (k-1)-subsets s of c do
ƒ if (s is not in Lk-1) then delete c from Ck
ƒ Use object-relational extensions like UDFs, BLOBs, and Table
functions for efficient implementation [S. Sarawagi, S. Thomas, and R.
Agrawal. Integrating association rule mining with relational database
systems: Alternatives and implications. SIGMOD’98]

23

Further Improvements of Mining Methods

ƒ AFOPT (Liu, et al. @ KDD’03)

ƒ A “push-right” method for mining condensed frequent pattern
(CFP) tree

ƒ Carpenter (Pan, et al. @ KDD’03)

ƒ Mine data sets with small rows but numerous columns
ƒ Construct a row-enumeration tree for efficient mining

ƒ FPgrowth+ (Grahne and Zhu, FIMI’03)

ƒ Efficiently Using Prefix-Trees in Mining Frequent Itemsets, Proc.
ICDM'03 Int. Workshop on Frequent Itemset Mining
Implementations (FIMI'03), Melbourne, FL, Nov. 2003

ƒ TD-Close (Liu, et al, SDM’06)

24
Further Improvement of the Apriori Method

ƒ Major computational challenges

ƒ Multiple scans of transaction database
ƒ Huge number of candidates
ƒ Tedious workload of support counting for candidates

ƒ Improving Apriori: general ideas

ƒ Reduce passes of transaction database scans
ƒ Shrink number of candidates
ƒ Facilitate support counting of candidates

25

Partition: Scan Database Only Twice

ƒ Any itemset that is potentially frequent in DB must

be frequent in at least one of the partitions of DB
ƒ Scan 1: partition database and find local frequent
patterns
ƒ Scan 2: consolidate global frequent patterns
ƒ A. Savasere, E. Omiecinski and S. Navathe,
VLDB’95

DB1 + DB2 + + DBk = DB

sup1(i) < σDB1 sup2(i) < σDB2 supk(i) < σDBk sup(i) < σDB

26
DHP: Reduce the Number of Candidates

ƒ A k-itemset whose corresponding hashing bucket

count is below the threshold cannot be frequent
ƒ Candidates: a, b, c, d, e
ƒ Hash entries: {ab, ad, ae} {bd, be, de} …
ƒ Frequent 1-itemset: a, b, d, e
ƒ ab is not a candidate 2-itemset if the sum of count of
{ab, ad, ae} is below support threshold

ƒ J. Park, M. Chen, and P. Yu. An effective hash-

based algorithm for mining association rules. In
SIGMOD’95
27

Sampling for Frequent Patterns

ƒ Select a sample of original database, mine

frequent patterns within sample using Apriori
ƒ Scan database once to verify frequent itemsets
found in sample, only borders of closure of
frequent patterns are checked
ƒ Example: check abcd instead of ab, ac, …, etc.

ƒ Scan database again to find missed frequent

patterns
ƒ H. Toivonen. Sampling large databases for
association rules. In VLDB’96
28
DIC: Reduce Number of Scans

ABCD
ƒ Once both A and D are determined
frequent, the counting of AD begins
ABC ABD ACD BCD ƒ Once all length-2 subsets of BCD are
determined frequent, the counting of
BCD begins
AB AC BC AD BD CD
Transactions
1-itemsets
A B C D
Apriori 2-itemsets
…
{}
Itemset lattice 1-itemsets
S. Brin R. Motwani, J. Ullman, 2-items
and S. Tsur. Dynamic itemset DIC 3-items
counting and implication rules for
market basket data. In
SIGMOD’97

29

Pattern-Growth Approach:
Mining Frequent Patterns Without Candidate Generation

ƒ Bottlenecks of the Apriori approach

ƒ Breadth-first (i.e., level-wise) search
ƒ Candidate generation and test
ƒ Often generates a huge number of candidates
ƒ The FPGrowth Approach (J. Han, J. Pei, and Y. Yin,
SIGMOD’ 00)
ƒ Depth-first search
ƒ Avoid explicit candidate generation
ƒ Major philosophy: Grow long patterns from short ones using
local frequent items only
ƒ “abc” is a frequent pattern
ƒ Get all transactions having “abc”, i.e., project DB on abc: DB|abc
ƒ “d” is a local frequent item in DB|abc Æ abcd is a frequent pattern
30
Construct FP-tree from a Transaction Database

TID Items bought (ordered) frequent items

100 {f, a, c, d, g, i, m, p} {f, c, a, m, p}
200 {a, b, c, f, l, m, o} {f, c, a, b, m}
300 {b, f, h, j, o, w} {f, b} min_support = 3
400 {b, c, k, s, p} {c, b, p}
500 {a, f, c, e, l, p, m, n} {f, c, a, m, p} {}
Header Table
1. Scan DB once, find
frequent 1-itemset (single Item frequency head f:4 c:1
item pattern) f 4
c 4 c:3 b:1 b:1
2. Sort frequent items in a 3
frequency descending b 3 a:3 p:1
order, f-list m 3
p 3
3. Scan DB again, construct m:2 b:1
FP-tree
F-list = f-c-a-b-m-p p:2 m:1

31

Partition Patterns and Databases

ƒ Frequent patterns can be partitioned into

subsets according to f-list
ƒ F-list = f-c-a-b-m-p
ƒ Patterns containing p
ƒ Patterns having m but no p
ƒ …
ƒ Patterns having c but no a nor b, m, p
ƒ Pattern f
ƒ Completeness and non-redundancy

32
Find Patterns Having P From P-conditional Database

ƒ Starting at the frequent item header table in the FP-

tree
ƒ Traverse the FP-tree by following the link of each
frequent item p
ƒ Accumulate all of transformed prefix paths of item p to
form p’s conditional pattern base
{}
Header Table
f:4 c:1 Conditional pattern bases
Item frequency head
f 4 item cond. pattern base
c 4 c:3 b:1 b:1 c f:3
a 3
b 3 a:3 p:1 a fc:3
m 3 b fca:1, f:1, c:1
p 3 m:2 b:1 m fca:2, fcab:1
p:2 m:1 p fcam:2, cb:1
33

From Conditional Pattern-bases to Conditional FP-trees

ƒ For each pattern-base

ƒ Accumulate the count for each item in the base
ƒ Construct the FP-tree for the frequent items of the
pattern base

m-conditional pattern base:

{} fca:2, fcab:1
Header Table
Item frequency head All frequent
f:4 c:1 patterns relate to m
f 4 {}
c 4 c:3 b:1 b:1 m,
¼
a 3 f:3 ¼ fm, cm, am,
b 3 a:3 p:1 fcm, fam, cam,
m 3 c:3 fcam
m:2 b:1
p 3
p:2 m:1 a:3
m-conditional FP-tree

34
Recursion: Mining Each Conditional FP-tree

{}

{} Cond. pattern base of “am”: (fc:3) f:3

c:3
f:3
am-conditional FP-tree
c:3 {}
Cond. pattern base of “cm”: (f:3)
a:3 f:3
m-conditional FP-tree
cm-conditional FP-tree

{}

Cond. pattern base of “cam”: (f:3) f:3

cam-conditional FP-tree

35

A Special Case: Single Prefix Path in FP-tree

ƒ Suppose a (conditional) FP-tree T has a shared

single prefix-path P
{} ƒ Mining can be decomposed into two parts
ƒ Reduction of the single prefix path into one node
a1:n1
ƒ Concatenation of the mining results of the two parts
a2:n2
a3:n3
{} r1

b1:m1 C1:k1 a1:n1

¼ r1 =
a2:n2
+ b1:m1 C1:k1

C2:k2 C3:k3
a3:n3 C2:k2 C3:k3

36
Benefits of the FP-tree Structure

ƒ Completeness
ƒ Preserve complete information for frequent pattern
mining
ƒ Never break a long pattern of any transaction
ƒ Compactness
ƒ Reduce irrelevant info—infrequent items are gone
ƒ Items in frequency descending order: the more
frequently occurring, the more likely to be shared
ƒ Never be larger than the original database (not count
node-links and the count field)

37

The Frequent Pattern Growth Mining Method

ƒ Idea: Frequent pattern growth

ƒ Recursively grow frequent patterns by pattern and
database partition
ƒ Method
ƒ For each frequent item, construct its conditional
pattern-base, and then its conditional FP-tree
ƒ Repeat the process on each newly created conditional
FP-tree
ƒ Until the resulting FP-tree is empty, or it contains only
one path—single path will generate all the
combinations of its sub-paths, each of which is a
frequent pattern

38
Scaling FP-growth by Database Projection

ƒ What about if FP-tree cannot fit in memory?

ƒ DB projection
ƒ First partition a database into a set of projected DBs
ƒ Then construct and mine FP-tree for each projected DB
ƒ Parallel projection vs. partition projection techniques
ƒ Parallel projection
ƒ Project the DB in parallel for each frequent item
ƒ Parallel projection is space costly
ƒ All the partitions can be processed in parallel
ƒ Partition projection
ƒ Partition the DB based on the ordered frequent items
ƒ Passing the unprocessed parts to the subsequent partitions

39

Partition-Based Projection

ƒ Parallel projection needs a lot Tran. DB

of disk space fcamp
fcabm
ƒ Partition projection saves it fb
cbp
fcamp

p-proj DB m-proj DB b-proj DB a-proj DB c-proj DB f-proj DB

fcam fcab f fc f …
cb fca cb … …
fcam fca …

am-proj DB cm-proj DB
fc f …
fc f
fc f

40
FP-Growth vs. Apriori: Scalability With the Support Threshold

100 Data set T25I20D10K

90 D1 FP-grow th runtime
D1 Apriori runtime
80

70
Run time(sec.)

60

50

40

30

20

10

0
0 0.5 1 1.5 2 2.5 3
Support threshold(%)

41

FP-Growth vs. Tree-Projection: Scalability with the Support Threshold

Data set T25I20D100K

140
D2 FP-growth
120 D2 TreeProjection

100
Runtime (sec.)

80

60

40

20

0
0 0.5 1 1.5 2
Support threshold (%)

42
Advantages of the Pattern Growth Approach

ƒ Divide-and-conquer:
ƒ Decompose both the mining task and DB according to the
frequent patterns obtained so far
ƒ Lead to focused search of smaller databases
ƒ Other factors
ƒ No candidate generation, no candidate test
ƒ Compressed database: FP-tree structure
ƒ No repeated scan of entire database
ƒ Basic ops: counting local freq items and building sub FP-tree, no
pattern search and matching
ƒ A good open-source implementation and refinement of
FPGrowth
ƒ FPGrowth+ (Grahne and J. Zhu, FIMI'03)
43

Extension of Pattern Growth Mining Methodology

ƒ Mining closed frequent itemsets and max-patterns

ƒ CLOSET (DMKD’00), FPclose, and FPMax (Grahne & Zhu, Fimi’03)
ƒ Mining sequential patterns
ƒ PrefixSpan (ICDE’01), CloSpan (SDM’03), BIDE (ICDE’04)
ƒ Mining graph patterns
ƒ gSpan (ICDM’02), CloseGraph (KDD’03)
ƒ Constraint-based mining of frequent patterns
ƒ Convertible constraints (ICDE’01), gPrune (PAKDD’03)
ƒ Computing iceberg data cubes with complex measures
ƒ H-tree, H-cubing, and Star-cubing (SIGMOD’01, VLDB’03)
ƒ Pattern-growth-based Clustering
ƒ MaPle (Pei, et al., ICDM’03)
ƒ Pattern-Growth-Based Classification
ƒ Mining frequent and discriminative patterns (Cheng, et al, ICDE’07)
44
MaxMiner: Mining Max-patterns

ƒ 1st scan: find frequent items Tid Items

10 A,B,C,D,E
ƒ A, B, C, D, E
20 B,C,D,E,
ƒ 2nd scan: find support for 30 A,C,D,F
ƒ AB, AC, AD, AE, ABCDE
ƒ BC, BD, BE, BCDE Potential
ƒ CD, CE, CDE, DE, max-patterns
ƒ Since BCDE is a max-pattern, no need to check
BCD, BDE, CDE in later scan
ƒ R. Bayardo. Efficiently mining long patterns from
databases. SIGMOD’98
45

Mining Frequent Closed Patterns: CLOSET

ƒ Flist: list of all frequent items in support ascending order

ƒ Flist: d-a-f-e-c Min_sup=2
ƒ Divide search space TID Items
10 a, c, d, e, f
ƒ Patterns having d 20 a, b, e
ƒ Patterns having d but no a, etc. 30 c, e, f
40 a, c, d, f
50 c, e, f
ƒ Find frequent closed pattern recursively
ƒ Every transaction having d also has cfa Æ cfad is a frequent closed
pattern
ƒ J. Pei, J. Han & R. Mao. CLOSET: An Efficient Algorithm for
Mining Frequent Closed Itemsets", DMKD'00.

46
CLOSET+: Mining Closed Itemsets by Pattern-Growth

ƒ Itemset merging: if Y appears in every occurrence of X, then Y

is merged with X
ƒ Sub-itemset pruning: if Y ‫ כ‬X, and sup(X) = sup(Y), X and all of
X’s descendants in the set enumeration tree can be pruned
ƒ Hybrid tree projection
ƒ Bottom-up physical tree-projection
ƒ Top-down pseudo tree-projection
ƒ Item skipping: if a local frequent item has the same support in
several header tables at different levels, one can prune it from
the header table at higher levels
ƒ Efficient subset checking

47

CHARM / ECLAT: Mining by Exploring Vertical Data Format

ƒ Vertical format: t(AB) = {T11, T25, …}

ƒ tid-list: list of trans.-ids containing an itemset
ƒ Deriving closed patterns based on vertical intersections
ƒ t(X) = t(Y): X and Y always happen together
ƒ t(X) ⊂ t(Y): transaction having X always has Y
ƒ Using diffset to accelerate mining
ƒ Only keep track of differences of tids
ƒ t(X) = {T1, T2, T3}, t(XY) = {T1, T3}
ƒ Diffset (XY, X) = {T2}
ƒ Eclat/MaxEclat (Zaki et al. @KDD’97), VIPER(P. Shenoy
et al.@SIGMOD’00), CHARM (Zaki & Hsiao@SDM’02)

48
Visualization of Association Rules: Plane Graph

49

Visualization of Association Rules: Rule Graph

50
Visualization of Association Rules (SGI/MineSet 3.0)

51

Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ From association to correlation analysis
ƒ Mining colossal patterns
ƒ Summary
52
Mining Various Kinds of Association Rules

ƒ Mining multilevel association

ƒ Mining multidimensional association

ƒ Mining quantitative association

ƒ Mining interesting correlation patterns

53

Mining Multiple-Level Association Rules

ƒ Items often form hierarchies

ƒ Flexible support settings
ƒ Items at the lower level are expected to have lower support
ƒ Exploration of shared multi-level mining (Agrawal &
Srikant@VLB’95, Han & Fu@VLDB’95)

uniform support reduced support

Level 1
Milk Level 1
min_sup = 5%
[support = 10%] min_sup = 5%

Level 2 2% Milk Skim Milk Level 2

min_sup = 5% [support = 6%] [support = 4%] min_sup = 3%

54
Multi-level Association: Redundancy Filtering

ƒ Some rules may be redundant due to “ancestor”

relationships between items
ƒ Example
ƒ milk ⇒ wheat bread [support = 8%, confidence = 70%]
ƒ 2% milk ⇒ wheat bread [support = 2%, confidence = 72%]
ƒ We say the first rule is an ancestor of the
second rule
ƒ A rule is redundant if its support is close to the
“expected” value, based on the rule’s ancestor

55

Mining Multi-Dimensional Association

ƒ Single-dimensional rules:
ƒ buys(X, “milk”) ⇒ buys(X, “bread”)

ƒ Multi-dimensional rules: ≥ 2 dimensions or predicates

ƒ Inter-dimension assoc. rules (no repeated predicates)
ƒ age(X,”19-25”) ∧ occupation(X,“student”) ⇒ buys(X, “coke”)
ƒ hybrid-dimension assoc. rules (repeated predicates)
ƒ age(X,”19-25”) ∧ buys(X, “popcorn”) ⇒ buys(X, “coke”)
ƒ Categorical Attributes: finite number of possible values, no
ordering among values—data cube approach
ƒ Quantitative Attributes: Numeric, implicit ordering among
values—discretization, clustering, and gradient approaches

56
Mining Quantitative Associations

ƒ Techniques can be categorized by how numerical

attributes, such as age or salary are treated:

1. Static discretization based on predefined concept hierarchies

(data cube methods)
2. Dynamic discretization based on data distribution (quantitative
rules, e.g., Agrawal & Srikant@SIGMOD96)
3. Clustering: Distance-based association (e.g., Yang &
Miller@SIGMOD97)
ƒ One dimensional clustering then association
4. Deviation: (such as Aumann and Lindell@KDD99)
Sex = female => Wage: mean=$7/hr (overall mean = $9)

57

Static Discretization of Quantitative Attributes

ƒ Discretized prior to mining using concept hierarchy.

ƒ Numeric values are replaced by ranges
ƒ In relational database, finding all frequent k-predicate sets
will require k or k+1 table scans
()
ƒ Data cube is well suited for mining
ƒ The cells of an n-dimensional (age) (income) (buys)
cuboid correspond to the
predicate sets
(age, income) (age,buys) (income,buys)
ƒ Mining from data cubes
can be much faster
(age,income,buys)

58
Quantitative Association Rules

ƒ Proposed by Lent, Swami and Widom ICDE’97

ƒ Numeric attributes are dynamically discretized
» Such that the confidence or compactness of the rules mined is
maximized
ƒ 2-D quantitative association
rules: Aquan1 ∧ Aquan2 ⇒ Acat
ƒ Cluster adjacent association
rules to form general rules
using a 2-D grid
ƒ Example
age(X, “34-35”) ∧ income(X,
“30-50K”) ⇒ buys(X, “high
resolution TV”)

59

Mining Other Interesting Patterns

ƒ Flexible support constraints (Wang, et al. @

VLDB’02)
ƒ Some items (e.g., diamond) may occur rarely but are
valuable
ƒ Customized supmin specification and application
ƒ Top-K closed frequent patterns (Han, et al. @
ICDM’02)
ƒ Hard to specify supmin, but top-k with lengthmin is more
desirable
ƒ Dynamically raise supmin in FP-tree construction and
mining, and select most promising path to mine

60
Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

61

Interestingness Measure: Correlations (Lift)

ƒ play basketball ⇒ eat cereal [40%, 66.7%] is misleading

ƒ The overall % of students eating cereal is 75% > 66.7%.

ƒ play basketball ⇒ not eat cereal [20%, 33.3%] is more

accurate, although with lower support and confidence
ƒ Measure of dependent/correlated events: lift
P( A∪ B) Basketball Not basketball Sum (row)
lift = Cereal 2000 1750 3750
P( A) P( B)
Not cereal 1000 250 1250
2000 / 5000
lift ( B, C ) = = 0.89 Sum(col.) 3000 2000 5000
3000 / 5000 * 3750 / 5000
1000 / 5000
lift ( B, ¬C ) = = 1.33
3000 / 5000 *1250 / 5000

62
Are lift and χ2 Good Measures of Correlation?

ƒ “Buy walnuts ⇒ buy

milk [1%, 80%]” is
misleading if 85% of
customers buy milk
ƒ Support and confidence
are not good to indicate
correlations
ƒ Over 20 interestingness
measures have been
proposed (see Tan,
Kumar, Sritastava
@KDD’02)
ƒ Which are good ones?

63

Null-Invariant Measures

64
Comparison of Interestingness Measures

ƒ Null-(transaction) invariance is crucial for correlation analysis

ƒ Lift and χ2 are not null-invariant
ƒ 5 null-invariant measures
Milk No Milk Sum
(row)
Coffee m, c ~m, c c
No m, ~c ~m, ~c ~c
Coffee
Sum(col. m ~m Σ
)

Null-transactions Kulczynski
w.r.t. m and c measure (1927) Null-invariant

Subtle: They disagree

65

Analysis of DBLP Coauthor Relationships

Recent DB conferences, removing balanced associations, low sup, etc.

Advisor-advisee relation: Kulc: high,

coherence: low, cosine: middle
ƒ Tianyi Wu, Yuguo Chen and Jiawei Han, “Association
Mining in Large Databases: A Re-Examination of Its
Measures”, Proc. 2007 Int. Conf. Principles and Practice of
Knowledge Discovery in Databases (PKDD'07), Sept. 2007
66
Which Null-Invariant Measure Is Better?

ƒ IR (Imbalance Ratio): measure the imbalance of two

itemsets A and B in rule implications

ƒ Kulczynski and Imbalance Ratio (IR) together present a

clear picture for all the three datasets D4 through D6
ƒ D4 is balanced & neutral
ƒ D5 is imbalanced & neutral
ƒ D6 is very imbalanced & neutral

67

Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

68
Constraint-based (Query-Directed) Mining

ƒ Finding all the patterns in a database autonomously? —

unrealistic!
ƒ The patterns could be too many but not focused!
ƒ Data mining should be an interactive process
ƒ User directs what to be mined using a data mining query
language (or a graphical user interface)
ƒ Constraint-based mining
ƒ User flexibility: provides constraints on what to be mined
ƒ System optimization: explores such constraints for efficient
mining — constraint-based mining: constraint-pushing, similar to
push selection first in DB query processing
ƒ Note: still find all the answers satisfying constraints, not finding
some answers in “heuristic search”
69

Constraints in Data Mining

ƒ Knowledge type constraint:

ƒ classification, association, etc.
ƒ Data constraint — using SQL-like queries
ƒ find product pairs sold together in stores in Chicago in
Dec.’02
ƒ Dimension/level constraint
ƒ in relevance to region, price, brand, customer category
ƒ Rule (or pattern) constraint
ƒ small sales (price < $10) triggers big sales (sum >
$200)
ƒ Interestingness constraint
ƒ strong rules: min_support ≥ 3%, min_confidence ≥
60%
70
Constraint-Based Frequent Pattern Mining

ƒ Classification of constraints based on their

constraint-pushing capabilities
ƒ Anti-monotonic: If constraint c is violated, its further
mining can be terminated
ƒ Monotonic: If c is satisfied, no need to check c again
ƒ Data anti-monotonic: If a transaction t does not satisfy
c, t can be pruned from its further mining
ƒ Succinct: c must be satisfied, so one can start with
the data sets satisfying c
ƒ Convertible: c is not monotonic nor anti-monotonic,
but it can be converted into it if items in the
transaction can be properly ordered

71

Anti-Monotonicity in Constraint Pushing

TDB (min_sup=2)
TID Transaction
ƒ A constraint C is antimonotone if the super
pattern satisfies C, all of its sub-patterns do so 10 a, b, c, d, f
too 20 b, c, d, f, g, h
ƒ In other words, anti-monotonicity: If an itemset S 30 a, c, d, e, f
violates the constraint, so does any of its 40 c, e, f, g
superset Item Profit
ƒ Ex. 1. sum(S.price) ≤ v is anti-monotone a 40
ƒ Ex. 2. range(S.profit) ≤ 15 is anti-monotone b 0
ƒ Itemset ab violates C c -20
ƒ So does every superset of ab
d 10
ƒ Ex. 3. sum(S.Price) ≥ v is not anti-monotone
e -30
ƒ Ex. 4. support count is anti-monotone: core
f 30
property used in Apriori
g 20
h -10
72
Monotonicity for Constraint Pushing

TDB (min_sup=2)
TID Transaction
ƒ A constraint C is monotone if the
10 a, b, c, d, f
pattern satisfies C, we do not need to
20 b, c, d, f, g, h
check C in subsequent mining
30 a, c, d, e, f
ƒ Alternatively, monotonicity: If an itemset 40 c, e, f, g
S satisfies the constraint, so does any Item Profit
of its superset a 40
b 0
ƒ Ex. 1. sum(S.Price) ≥ v is monotone
c -20
ƒ Ex. 2. min(S.Price) ≤ v is monotone d 10
ƒ Ex. 3. C: range(S.profit) ≥ 15 e -30
ƒ Itemset ab satisfies C f 30
ƒ So does every superset of ab g 20
h -10
73

Data Antimonotonicity: Pruning Data Space

TDB (min_sup=2)
TID Transaction
ƒ A constraint c is data antimonotone if for a pattern 10 a, b, c, d, f, h
p cannot satisfy a transaction t under c, p’s
20 b, c, d, f, g, h
superset cannot satisfy t under c either
30 b, c, d, f, g
ƒ The key for data antimonotone is recursive data
40 c, e, f, g
reduction Item Profit
ƒ Ex. 1. sum(S.Price) ≥ v is data antimonotone a 40
ƒ Ex. 2. min(S.Price) ≤ v is data antimonotone b 0
ƒ Ex. 3. C: range(S.profit) ≥ 25 is data antimonotone c -20
ƒ Itemset {b, c}’s projected DB: d -15
ƒ T10’: {d, f, h}, T20’: {d, f, g, h}, T30’: {d, f, g} e -30
ƒ since C cannot satisfy T10’, T10’ can be pruned f -10
g 20
h -5
74
Succinctness

ƒ Succinctness:
ƒ Given A1, the set of items satisfying a succinctness
constraint C, then any set S satisfying C is based on
A1 , i.e., S contains a subset belonging to A1
ƒ Idea: Without looking at the transaction database,
whether an itemset S satisfies constraint C can be
determined based on the selection of items
ƒ min(S.Price) ≤ v is succinct
ƒ sum(S.Price) ≥ v is not succinct
ƒ Optimization: If C is succinct, C is pre-counting
pushable
75

The Apriori Algorithm — Example

Database D itemset sup.

L1 itemset sup.
TID Items C1 {1} 2 {1} 2
100 134 {2} 3 {2} 3
200 235 Scan D {3} 3 {3} 3
300 1235 {4} 1 {5} 3
400 25 {5} 3
C2 itemset sup C2 itemset
L2 itemset sup {1 2} 1 Scan D {1 2}
{1 3} 2 {1 3} 2 {1 3}
{2 3} 2 {1 5} 1 {1 5}
{2 3} 2 {2 3}
{2 5} 3
{2 5} 3 {2 5}
{3 5} 2
{3 5} 2 {3 5}
C3 itemset Scan D L3 itemset sup
{2 3 5} {2 3 5} 2
76
Naïve Algorithm: Apriori + Constraint

Database D itemset sup.

L1 itemset sup.
TID Items C1 {1} 2 {1} 2
100 134 {2} 3 {2} 3
200 235 Scan D {3} 3 {3} 3
300 1235 {4} 1 {5} 3
400 25 {5} 3
C2 itemset sup C2 itemset
L2 itemset sup {1 2} 1 Scan D {1 2}
{1 3} 2 {1 3} 2 {1 3}
{2 3} 2 {1 5} 1 {1 5}
{2 3} 2 {2 3}
{2 5} 3
{2 5} 3 {2 5}
{3 5} 2
{3 5} 2 {3 5}
C3 itemset Scan D L3 itemset sup Constraint:
{2 3 5} {2 3 5} 2 Sum{S.price} < 5

77

The Constrained Apriori Algorithm: Push a Succinct Constraint Deep

Database D itemset sup.

L1 itemset sup.
TID Items C1 {1} 2 {1} 2
100 134 {2} 3 {2} 3
200 235 Scan D {3} 3 {3} 3
300 1235 {4} 1 {5} 3
400 25 {5} 3
C2 itemset sup C2 itemset
L2 itemset sup {1 2}
{1 2} 1 Scan D
{1 3} 2 {1 3} 2 {1 3}
not immediately
{1 5} 1 {1 5} to be used
{2 3} 2
{2 3} 2 {2 3}
{2 5} 3
{2 5} 3 {2 5}
{3 5} 2 {3 5}
{3 5} 2
C3 itemset Scan D L3 itemset sup Constraint:
{2 3 5} {2 3 5} 2 min{S.price } <= 1

78
The Constrained FP-Growth Algorithm: Push a Succinct Constraint Deep

TID Items TID Items

100 134 100 13
200 235
Remove
200 235
FP-Tree
300 1235 infrequent 300 1235
400 25 length 1
400 25

1-Projected DB
TID Items
100 3 4 No Need to project on 2, 3, or 5
300 2 3 5
Constraint:
min{S.price } <= 1

79

The Constrained FP-Growth Algorithm:

Push a Data Antimonotonic Constraint Deep

Remove from data

TID Items TID Items
100 134 100 1 3
200 235 300 1 3
FP-Tree
300 1235
400 25

Single branch, we are done

Constraint:
min{S.price } <= 1

80
The Constrained FP-Growth Algorithm: TID Transaction
Push a Data Antimonotonic Constraint Deep
10 a, b, c, d, f, h
20 b, c, d, f, g, h
30 b, c, d, f, g
TID Transaction 40 a, c, e, f, g
10 a, b, c, d, f,
h f, g, Item Profit
20 b, c, d,
FP-Tree a 40
30 b, c, hd, f, g
b 0
40 a, c, e, f, g
c -20
B-Projected DB Recursive
Data d -15
TID Transaction Pruning
e -30
10 a, c, d, f, h
20 c, d, f, g, h B f -10
FP-Tree g 20
30 c, d, f, g
h -5

Single branch: Constraint:

range{S.price } > 25
bcdfg: 2 min_sup >= 2
81

Converting “Tough” Constraints

ƒ Convert tough constraints into TDB (min_sup=2)

TID Transaction
anti-monotone or monotone by
10 a, b, c, d, f
properly ordering items 20 b, c, d, f, g, h
ƒ Examine C: avg(S.profit) ≥ 25 30 a, c, d, e, f
40 c, e, f, g
ƒ Order items in value-descending
Item Profit
order
a 40
ƒ <a, f, g, d, b, h, c, e> b 0
ƒ If an itemset afb violates C c -20
d 10
ƒ So does afbh, afb* e -30
ƒ It becomes anti-monotone! f 30
g 20
h -10
82
Strongly Convertible Constraints

ƒ avg(X) ≥ 25 is convertible anti-monotone

w.r.t. item value descending order R: <a,
f, g, d, b, h, c, e>
ƒ If an itemset af violates a constraint C, so Item Profit
does every itemset with af as prefix, such as a 40
afd b 0
ƒ avg(X) ≥ 25 is convertible monotone c -20
w.r.t. item value ascending order R-1: d 10
<e, c, h, b, d, g, f, a> e -30
f 30
ƒ If an itemset d satisfies a constraint C, so
does itemsets df and dfa, which having d as g 20
a prefix h -10

ƒ Thus, avg(X) ≥ 25 is strongly convertible

83

Can Apriori Handle Convertible Constraints?

ƒ A convertible, not monotone nor anti-

monotone nor succinct constraint cannot
Item Value
be pushed deep into the an Apriori mining a 40
algorithm b 0
ƒ Within the level wise framework, no direct c -20
pruning based on the constraint can be made d 10

ƒ Itemset df violates constraint C: avg(X) >= 25 e -30

f 30
ƒ Since adf satisfies C, Apriori needs df to
g 20
assemble adf, df cannot be pruned
h -10
ƒ But it can be pushed into frequent-pattern
growth framework!
84
Mining With Convertible Constraints

ƒ C: avg(X) >= 25, min_sup=2 Item Value

ƒ List items in every transaction in value a 40
descending order R: <a, f, g, d, b, h, c, e> f 30
ƒ C is convertible anti-monotone w.r.t. R g 20
d 10
ƒ Scan TDB once
b 0
ƒ remove infrequent items h -10
ƒ Item h is dropped c -20
ƒ Itemsets a and f are good, … e -30
TDB (min_sup=2)
ƒ Projection-based mining
TID Transaction
ƒ Imposing an appropriate order on item
10 a, f, d, b, c
projection
20 f, g, d, b, c
ƒ Many tough constraints can be converted into
30 a, f, d, c, e
(anti)-monotone
40 f, g, h, c, e
85

Handling Multiple Constraints

ƒ Different constraints may require different or

even conflicting item-ordering
ƒ If there exists an order R s.t. both C1 and C2
are convertible w.r.t. R, then there is no
conflict between the two convertible
constraints
ƒ If there exists conflict on order of items
ƒ Try to satisfy one constraint first
ƒ Then using the order for the other constraint to
mine frequent itemsets in the corresponding
projected database
86
What Constraints Are Convertible?

Convertible Convertible Strongly

Constraint anti-monotone monotone convertible
avg(S) ≤ , ≥ v Yes Yes Yes
median(S) ≤ , ≥ v Yes Yes Yes
sum(S) ≤ v (items could be of any
Yes No No
value, v ≥ 0)
sum(S) ≤ v (items could be of any
No Yes No
value, v ≤ 0)
sum(S) ≥ v (items could be of any
No Yes No
value, v ≥ 0)
sum(S) ≥ v (items could be of any
Yes No No
value, v ≤ 0)
……

87

Constraint-Based Mining — A General Picture

Constraint Antimonotone Monotone Succinct

v∈S no yes yes
S⊇V no yes yes
S⊆V yes no yes
min(S) ≤ v no yes yes
min(S) ≥ v yes no yes
max(S) ≤ v yes no yes
max(S) ≥ v no yes yes
count(S) ≤ v yes no weakly
count(S) ≥ v no yes weakly
sum(S) ≤ v ( a ∈ S, a ≥ 0 ) yes no no
sum(S) ≥ v ( a ∈ S, a ≥ 0 ) no yes no
range(S) ≤ v yes no no
range(S) ≥ v no yes no
avg(S) θ v, θ ∈ { =, ≤, ≥ } convertible convertible no
support(S) ≥ ξ yes no no
support(S) ≤ ξ no yes no

88
A Classification of Constraints

Monotone
Antimonotone

Strongly
convertible
Succinct

Convertible Convertible
anti-monotone monotone

Inconvertible

89

Mining Frequent Patterns, Association and Correlations – Sub-Topics

ƒ Basic concepts and a road map

ƒ Scalable frequent itemset mining methods
ƒ Mining various kinds of association rules
ƒ From association to correlation analysis
ƒ Constraint-based association mining
ƒ Mining colossal patterns
ƒ Summary

90
Why Mining Colossal Frequent Patterns?

ƒ F. Zhu, X. Yan, J. Han, P. S. Yu, and H. Cheng, “Mining Colossal

Frequent Patterns by Core Pattern Fusion”, ICDE'07.
ƒ We have many algorithms, but can we mine large (i.e., colossal)
patterns? ― such as just size around 50 to 100? Unfortunately, not!
ƒ Why not? ― the curse of “downward closure” of frequent patterns
ƒ The “downward closure” property
ƒ Any sub-pattern of a frequent pattern is frequent.
ƒ Example. If (a1, a2, …, a100) is frequent, then a1, a2, …, a100, (a1, a2),
(a1, a3), …, (a1, a100), (a1, a2, a3), … are all frequent! There are about
2100 such frequent itemsets!
ƒ No matter using breadth-first search (e.g., Apriori) or depth-first search
(FPgrowth), we have to examine so many patterns
ƒ Thus the downward closure property leads to explosion!

91

Colossal Patterns: A Motivating Example

Let’s make a set of 40 transactions Closed/maximal patterns may

T1 = 1 2 3 4 ….. 39 40 partially alleviate the problem but not
T2 = 1 2 3 4 ….. 39 40 really solve it: We often need to mine
scattered large patterns!
: .
: . Let the minimum support threshold
: . σ= 20
40
 
: . There are 20 frequent patterns of
T40=1 2 3 4 ….. 39 40 size 20

Then delete the items on the diagonal Each is closed and maximal
# patterns = n 2n
T1 = 2 3 4 ….. 39 40   ≈ 2/π
T2 = 1 3 4 ….. 39 40
n/ 2 n
: .
The size of the answer set is
: .
exponential to n
: .
: .
T40=1 2 3 4 …… 39 92
Colossal Pattern Set: Small but Interesting

ƒ It is often the case that

only a small number of
patterns are colossal,
i.e., of large size

ƒ Colossal patterns are

usually attached with
greater importance than
those of small pattern
sizes

93

Mining Colossal Patterns: Motivation and Philosophy

ƒ Motivation: Many real-world tasks need mining colossal patterns

ƒ Micro-array analysis in bioinformatics (when support is low)
ƒ Biological sequence patterns
ƒ Biological/sociological/information graph pattern mining
ƒ No hope for completeness
ƒ If the mining of mid-sized patterns is explosive in size, there is no hope to
find colossal patterns efficiently by insisting “complete set” mining
philosophy
ƒ Jumping out of the swamp of the mid-sized results
ƒ What we may develop is a philosophy that may jump out of the swamp of
mid-sized results that are explosive in size and jump to reach colossal
patterns
ƒ Striving for mining almost complete colossal patterns
ƒ The key is to develop a mechanism that may quickly reach colossal
patterns and discover most of them

94
Alas, A Show of Colossal Pattern Mining!

Let the min-support threshold σ= 20

 40 
T1 = 2 3 4 ….. 39 40 Then there are  20  closed/maximal
T2 = 1 3 4 ….. 39 40 frequent patterns of size 20
: .
: . However, there is only one with size
: . greater than 20, (i.e., colossal):
: .
α= {41,42,…,79} of size 39
T40=1 2 3 4 …… 39
T41= 41 42 43 ….. 79
T42= 41 42 43 ….. 79
: .
The
The existing
existing fastest
fastest mining
mining algorithms
algorithms
: . (e.g.,
(e.g., FPClose, LCM)fail
FPClose, LCM) failto
tocomplete
complete
T60= 41 42 43 … 79 running
running
The
Thealgorithm
algorithmoutputs
outputsthis
thiscolossal
colossal
pattern in seconds
pattern in seconds
95

Methodology of Pattern-Fusion Strategy

ƒ Pattern-Fusion traverses the tree in a bounded-breadth

way
ƒ Always pushes down a frontier of a bounded-size candidate
pool
ƒ Only a fixed number of patterns in the current candidate pool
will be used as the starting nodes to go down in the pattern tree
― thus avoids the exponential search space

ƒ Pattern-Fusion identifies “shortcuts” whenever possible

ƒ Pattern growth is not performed by single-item addition but by
leaps and bounded: agglomeration of multiple patterns in the
pool
ƒ These shortcuts will direct the search down the tree much more
rapidly towards the colossal patterns
96
Observation: Colossal Patterns and Core Patterns

Transaction Database D
A colossal pattern α
α D

α1 Dαk
α2
D
Dα1
α
Dα2

αk

Subpatterns α1 to αk cluster tightly around the colossal pattern α by

sharing a similar support. We call such subpatterns core patterns of α

97

Robustness of Colossal Patterns

ƒ Core Patterns
Intuitively, for a frequent pattern α, a subpattern β is a τ-core
pattern of α if β shares a similar support set with α, i.e.,
| Dα |
≥τ 0 <τ ≤1
| Dβ |
where τ is called the core ratio

ƒ Robustness of Colossal Patterns

A colossal pattern is robust in the sense that it tends to have much
more core patterns than small patterns

98
Example: Core Patterns

ƒ A colossal pattern has far more core patterns than a small-sized pattern
ƒ A colossal pattern has far more core descendants of a smaller size c
ƒ A random draw from a complete set of pattern of size c would more
likely to pick a core descendant of a colossal pattern
ƒ A colossal pattern can be generated by merging a set of core patterns

Transaction (# of Ts) Core Patterns (τ = 0.5)

(abe) (100) (abe), (ab), (be), (ae), (e)
(bcf) (100) (bcf), (bc), (bf)
(acf) (100) (acf), (ac), (af)

(abcef) (100) (ab), (ac), (af), (ae), (bc), (bf), (be) (ce), (fe), (e), (abc),
(abf), (abe), (ace), (acf), (afe), (bcf), (bce), (bfe), (cfe),
(abcf), (abce), (bcfe), (acfe), (abfe), (abcef)

99

Robustness of Colossal Patterns

ƒ (d,τ)-robustness: A pattern α is (d, τ)-robust if d is the maximum

number of items that can be removed from α for the resulting pattern
to remain a τ-core pattern of α
ƒ For a (d,τ)-robust pattern α, it has Ω(2d ) core patterns
» A colossal patterns tend to have a large number of core patterns
ƒ Pattern distance: For patterns α and β, the pattern distance of α and β
is defined to be Dα ∩ Dβ
Dist(α , β ) = 1 −
Dα ∪ Dβ
ƒ If two patterns α and β are both core patterns of a same pattern, they
would be bounded by a “ball” of a radius specified by their core ratio τ

1
Dist ( α , β ) ≤ 1 − = r (τ )
2 /τ − 1
ƒ Once we identify one core pattern, we will be able to find all the other
core patterns by a bounding ball of radius r(τ)
100
Colossal Patterns Correspond to Dense Balls

ƒ Due to their robustness,

colossal patterns
correspond to dense balls
ƒ Ω( 2^d) in population
ƒ A random draw in the
pattern space will hit
somewhere in the ball with
high probability

101

Idea of Pattern-Fusion Algorithm

ƒ Generate a complete set of frequent patterns up

to a small size
ƒ Randomly pick a pattern β, and β has a high
probability to be a core-descendant of some
colossal pattern α
ƒ Identify all α’s descendants in this complete set,
and merge all of them ― This would generate a
much larger core-descendant of α
ƒ In the same fashion, we select K patterns. This
set of larger core-descendants will be the
candidate pool for the next iteration 102
Pattern-Fusion: The Algorithm

ƒ Initialization (Initial pool): Use an existing algorithm to

mine all frequent patterns up to a small size, e.g., 3
ƒ Iteration (Iterative Pattern Fusion):
ƒ At each iteration, k seed patterns are randomly picked from the
current pattern pool
ƒ For each seed pattern thus picked, we find all the patterns within a
bounding ball centered at the seed pattern
ƒ All these patterns found are fused together to generate a set of
super-patterns. All the super-patterns thus generated form a new
pool for the next iteration
ƒ Termination: when the current pool contains no more than
K patterns at the beginning of an iteration

103

Why Is Pattern-Fusion Efficient?

ƒ A bounded-breadth
pattern tree traversal
ƒ It avoids explosion in
mining mid-sized ones
ƒ Randomness comes to
help to stay on the right
path
ƒ Ability to identify “short-
cuts” and take “leaps”
ƒ fuse small patterns
together in one step to
generate new patterns of
significant sizes
ƒ Efficiency
104
Pattern-Fusion Leads to Good Approximation

ƒ Gearing toward colossal patterns

ƒ The larger the pattern, the greater the chance it will
be generated
ƒ Catching outliers
ƒ The more distinct the pattern, the greater the chance
it will be generated

105

Experimental Setting

ƒ Synthetic data set

ƒ Diagn an n x (n-1) table where ith row has integers from 1 to n except
i. Each row is taken as an itemset. min_support is n/2.
ƒ Real data set
ƒ Replace: A program trace data set collected from the “replace”
program, widely used in software engineering research
ƒ ALL: A popular gene expression data set, a clinical data on ALL-AML
leukemia (www.broad.mit.edu/tools/data.html).
ƒ Each item is a column, representing the activitiy level of
gene/protein in the same
ƒ Frequent pattern would reveal important correlation between
gene expression patterns and disease outcomes

106
Experiment Results on Diagn

ƒ LCM run time increases

exponentially with pattern
size n
ƒ Pattern-Fusion finishes
efficiently
ƒ The approximation error
of Pattern-Fusion (with
min-sup 20) in
comparison with the
complete set) is rather
close to uniform sampling
(which randomly picks K
patterns from the
complete answer set)
107

Experimental Results on ALL

ƒ ALL: A popular gene expression data set with 38

transactions, each with 866 columns
ƒ There are 1736 items in total
ƒ The table shows a high frequency threshold of 30

108
Experimental Results on REPLACE

ƒ REPLACE
ƒ A program trace data set, recording 4395
calls and transitions
ƒ The data set contains 4395 transactions with
57 items in total
ƒ With support threshold of 0.03, the largest
patterns are of size 44
ƒ They are all discovered by Pattern-Fusion
with different settings of K and τ, when started
with an initial pool of 20948 patterns of size
<=3

109

Experimental Results on REPLACE

ƒ Approximation error when

compared with the complete
mining result 0.01
K=50
ƒ Example. Out of the total 98 0.009 K=100
K=200
Approximation Error ∆(AQ)

0.008
P

patterns of size >=42, when 0.007

K=100, Pattern-Fusion returns 0.006

80 of them 0.005

0.004
ƒ A good approximation to the
0.003
colossal patterns in the sense 0.002
that any pattern in the complete 0.001

set is on average at most 0.17 0

39 40 41 42 43 44 45
items away from one of these Pattern Size (>=)

80 patterns

110
Agenda

11 Session
Session Overview
Overview

Mining
Mining Frequent
Frequent Patterns,
Patterns,
22 Association,
Association, and
and Correlations
Correlations

33 Summary
Summary and
and Conclusion
Conclusion

111

Frequent-Pattern Mining: Summary

ƒ Frequent pattern mining—an important task in data

mining
ƒ Scalable frequent pattern mining methods
ƒ Apriori (Candidate generation & test)
ƒ Projection-based (FPgrowth, CLOSET+, ...)
ƒ Vertical format approach (CHARM, ...)

ƒ Mining a variety of rules and interesting patterns

ƒ Constraint-based mining
ƒ Mining sequential and structured patterns
ƒ Extensions and applications 112
Frequent-Pattern Mining: Research Problems

ƒ Mining fault-tolerant frequent, sequential and

structured patterns
ƒ Patterns allows limited faults (insertion, deletion,
mutation)
ƒ Mining truly interesting patterns
ƒ Surprising, novel, concise, …
ƒ Application exploration
ƒ E.g., DNA sequence analysis and bio-pattern
classification
ƒ “Invisible” data mining

113

Ref: Basic Concepts of Frequent Pattern Mining

ƒ (Association Rules) R. Agrawal, T. Imielinski, and A.

Swami. Mining association rules between sets of items
in large databases. SIGMOD'93.
ƒ (Max-pattern) R. J. Bayardo. Efficiently mining long
patterns from databases. SIGMOD'98.
ƒ (Closed-pattern) N. Pasquier, Y. Bastide, R. Taouil, and
L. Lakhal. Discovering frequent closed itemsets for
association rules. ICDT'99.
ƒ (Sequential pattern) R. Agrawal and R. Srikant. Mining
sequential patterns. ICDE'95

114
Ref: Apriori and Its Improvements

ƒ R. Agrawal and R. Srikant. Fast algorithms for mining association rules.

VLDB'94.
ƒ H. Mannila, H. Toivonen, and A. I. Verkamo. Efficient algorithms for
discovering association rules. KDD'94.
ƒ A. Savasere, E. Omiecinski, and S. Navathe. An efficient algorithm for
mining association rules in large databases. VLDB'95.
ƒ J. S. Park, M. S. Chen, and P. S. Yu. An effective hash-based algorithm
for mining association rules. SIGMOD'95.
ƒ H. Toivonen. Sampling large databases for association rules. VLDB'96.
ƒ S. Brin, R. Motwani, J. D. Ullman, and S. Tsur. Dynamic itemset
counting and implication rules for market basket analysis. SIGMOD'97.
ƒ S. Sarawagi, S. Thomas, and R. Agrawal. Integrating association rule
mining with relational database systems: Alternatives and implications.
SIGMOD'98.

115

Ref: Depth-First, Projection-Based FP Mining

ƒ R. Agarwal, C. Aggarwal, and V. V. V. Prasad. A tree projection algorithm for

generation of frequent itemsets. J. Parallel and Distributed Computing:02.
ƒ J. Han, J. Pei, and Y. Yin. Mining frequent patterns without candidate
generation. SIGMOD’ 00.
ƒ J. Liu, Y. Pan, K. Wang, and J. Han. Mining Frequent Item Sets by
Opportunistic Projection. KDD'02.
ƒ J. Han, J. Wang, Y. Lu, and P. Tzvetkov. Mining Top-K Frequent Closed
Patterns without Minimum Support. ICDM'02.
ƒ J. Wang, J. Han, and J. Pei. CLOSET+: Searching for the Best Strategies for
Mining Frequent Closed Itemsets. KDD'03.
ƒ G. Liu, H. Lu, W. Lou, J. X. Yu. On Computing, Storing and Querying Frequent
Patterns. KDD'03.
ƒ G. Grahne and J. Zhu, Efficiently Using Prefix-Trees in Mining Frequent
Itemsets, Proc. ICDM'03 Int. Workshop on Frequent Itemset Mining
Implementations (FIMI'03), Melbourne, FL, Nov. 2003

116
Ref: Vertical Format and Row Enumeration Methods

ƒ M. J. Zaki, S. Parthasarathy, M. Ogihara, and W. Li. Parallel algorithm

for discovery of association rules. DAMI:97.
ƒ Zaki and Hsiao. CHARM: An Efficient Algorithm for Closed Itemset
Mining, SDM'02.
ƒ C. Bucila, J. Gehrke, D. Kifer, and W. White. DualMiner: A Dual-
Pruning Algorithm for Itemsets with Constraints. KDD’02.
ƒ F. Pan, G. Cong, A. K. H. Tung, J. Yang, and M. Zaki , CARPENTER:
Finding Closed Patterns in Long Biological Datasets. KDD'03.
ƒ H. Liu, J. Han, D. Xin, and Z. Shao, Mining Interesting Patterns from
Very High Dimensional Data: A Top-Down Row Enumeration
Approach, SDM'06.

117

Ref: Mining Multi-Level and Quantitative Rules

ƒ R. Srikant and R. Agrawal. Mining generalized association rules.

VLDB'95.
ƒ J. Han and Y. Fu. Discovery of multiple-level association rules from
large databases. VLDB'95.
ƒ R. Srikant and R. Agrawal. Mining quantitative association rules in
large relational tables. SIGMOD'96.
ƒ T. Fukuda, Y. Morimoto, S. Morishita, and T. Tokuyama. Data mining
using two-dimensional optimized association rules: Scheme,
algorithms, and visualization. SIGMOD'96.
ƒ K. Yoda, T. Fukuda, Y. Morimoto, S. Morishita, and T. Tokuyama.
Computing optimized rectilinear regions for association rules. KDD'97.
ƒ R.J. Miller and Y. Yang. Association rules over interval data.
SIGMOD'97.
ƒ Y. Aumann and Y. Lindell. A Statistical Theory for Quantitative
Association Rules KDD'99.

118
Ref: Mining Correlations and Interesting Rules

ƒ M. Klemettinen, H. Mannila, P. Ronkainen, H. Toivonen, and A. I.

Verkamo. Finding interesting rules from large sets of discovered
association rules. CIKM'94.
ƒ S. Brin, R. Motwani, and C. Silverstein. Beyond market basket:
Generalizing association rules to correlations. SIGMOD'97.
ƒ C. Silverstein, S. Brin, R. Motwani, and J. Ullman. Scalable
techniques for mining causal structures. VLDB'98.
ƒ P.-N. Tan, V. Kumar, and J. Srivastava. Selecting the Right
Interestingness Measure for Association Patterns. KDD'02.
ƒ E. Omiecinski. Alternative Interest Measures for Mining
Associations. TKDE’03.
ƒ T. Wu, Y. Chen and J. Han, “Association Mining in Large Databases:
A Re-Examination of Its Measures”, PKDD'07

119

Ref: Mining Other Kinds of Rules

ƒ R. Meo, G. Psaila, and S. Ceri. A new SQL-like operator for mining

association rules. VLDB'96.
ƒ B. Lent, A. Swami, and J. Widom. Clustering association rules.
ICDE'97.
ƒ A. Savasere, E. Omiecinski, and S. Navathe. Mining for strong
negative associations in a large database of customer transactions.
ICDE'98.
ƒ D. Tsur, J. D. Ullman, S. Abitboul, C. Clifton, R. Motwani, and S.
Nestorov. Query flocks: A generalization of association-rule mining.
SIGMOD'98.
ƒ F. Korn, A. Labrinidis, Y. Kotidis, and C. Faloutsos. Ratio rules: A new
paradigm for fast, quantifiable data mining. VLDB'98.
ƒ F. Zhu, X. Yan, J. Han, P. S. Yu, and H. Cheng, “Mining Colossal
Frequent Patterns by Core Pattern Fusion”, ICDE'07.

120
Ref: Constraint-Based Pattern Mining

ƒ R. Srikant, Q. Vu, and R. Agrawal. Mining association rules with item

constraints. KDD'97
ƒ R. Ng, L.V.S. Lakshmanan, J. Han & A. Pang. Exploratory mining and
pruning optimizations of constrained association rules. SIGMOD’98
ƒ G. Grahne, L. Lakshmanan, and X. Wang. Efficient mining of
constrained correlated sets. ICDE'00
ƒ J. Pei, J. Han, and L. V. S. Lakshmanan. Mining Frequent Itemsets
with Convertible Constraints. ICDE'01
ƒ J. Pei, J. Han, and W. Wang, Mining Sequential Patterns with
Constraints in Large Databases, CIKM'02
ƒ F. Bonchi, F. Giannotti, A. Mazzanti, and D. Pedreschi. ExAnte:
Anticipated Data Reduction in Constrained Pattern Mining, PKDD'03
ƒ F. Zhu, X. Yan, J. Han, and P. S. Yu, “gPrune: A Constraint Pushing
Framework for Graph Pattern Mining”, PAKDD'07

121

Ref: Mining Sequential and Structured Patterns

ƒ R. Srikant and R. Agrawal. Mining sequential patterns: Generalizations

and performance improvements. EDBT’96.
ƒ H. Mannila, H Toivonen, and A. I. Verkamo. Discovery of frequent
episodes in event sequences. DAMI:97.
ƒ M. Zaki. SPADE: An Efficient Algorithm for Mining Frequent Sequences.
Machine Learning:01.
ƒ J. Pei, J. Han, H. Pinto, Q. Chen, U. Dayal, and M.-C. Hsu. PrefixSpan:
Mining Sequential Patterns Efficiently by Prefix-Projected Pattern
Growth. ICDE'01.
ƒ M. Kuramochi and G. Karypis. Frequent Subgraph Discovery.
ICDM'01.
ƒ X. Yan, J. Han, and R. Afshar. CloSpan: Mining Closed Sequential
Patterns in Large Datasets. SDM'03.
ƒ X. Yan and J. Han. CloseGraph: Mining Closed Frequent Graph
Patterns. KDD'03.

122
Ref: Mining Spatial, Multimedia, and Web Data

ƒ K. Koperski and J. Han, Discovery of Spatial Association

Rules in Geographic Information Databases, SSD’95.
ƒ O. R. Zaiane, M. Xin, J. Han, Discovering Web Access
Patterns and Trends by Applying OLAP and Data Mining
Technology on Web Logs. ADL'98.
ƒ O. R. Zaiane, J. Han, and H. Zhu, Mining Recurrent Items
in Multimedia with Progressive Resolution Refinement.
ICDE'00.
ƒ D. Gunopulos and I. Tsoukatos. Efficient Mining of
Spatiotemporal Patterns. SSTD'01.

123

Ref: Mining Frequent Patterns in Time-Series Data

ƒ B. Ozden, S. Ramaswamy, and A. Silberschatz. Cyclic association

rules. ICDE'98.
ƒ J. Han, G. Dong and Y. Yin, Efficient Mining of Partial Periodic Patterns
in Time Series Database, ICDE'99.
ƒ H. Lu, L. Feng, and J. Han. Beyond Intra-Transaction Association
Analysis: Mining Multi-Dimensional Inter-Transaction Association Rules.
TOIS:00.
ƒ B.-K. Yi, N. Sidiropoulos, T. Johnson, H. V. Jagadish, C. Faloutsos, and
A. Biliris. Online Data Mining for Co-Evolving Time Sequences.
ICDE'00.
ƒ W. Wang, J. Yang, R. Muntz. TAR: Temporal Association Rules on
Evolving Numerical Attributes. ICDE’01.
ƒ J. Yang, W. Wang, P. S. Yu. Mining Asynchronous Periodic Patterns in
Time Series Data. TKDE’03.
124
Ref: Iceberg Cube and Cube Computation

ƒ S. Agarwal, R. Agrawal, P. M. Deshpande, A. Gupta, J. F.

Naughton, R. Ramakrishnan, and S. Sarawagi. On the
computation of multidimensional aggregates. VLDB'96.
ƒ Y. Zhao, P. M. Deshpande, and J. F. Naughton. An array-based
algorithm for simultaneous multidi-mensional aggregates.
SIGMOD'97.
ƒ J. Gray, et al. Data cube: A relational aggregation operator
generalizing group-by, cross-tab and sub-totals. DAMI: 97.
ƒ M. Fang, N. Shivakumar, H. Garcia-Molina, R. Motwani, and J. D.
Ullman. Computing iceberg queries efficiently. VLDB'98.
ƒ S. Sarawagi, R. Agrawal, and N. Megiddo. Discovery-driven
exploration of OLAP data cubes. EDBT'98.
ƒ K. Beyer and R. Ramakrishnan. Bottom-up computation of sparse
and iceberg cubes. SIGMOD'99.

125

Ref: Iceberg Cube and Cube Exploration

ƒ J. Han, J. Pei, G. Dong, and K. Wang, Computing Iceberg Data

Cubes with Complex Measures. SIGMOD’ 01.
ƒ W. Wang, H. Lu, J. Feng, and J. X. Yu. Condensed Cube: An
Effective Approach to Reducing Data Cube Size. ICDE'02.
ƒ G. Dong, J. Han, J. Lam, J. Pei, and K. Wang. Mining Multi-
Dimensional Constrained Gradients in Data Cubes. VLDB'01.
ƒ T. Imielinski, L. Khachiyan, and A. Abdulghani. Cubegrades:
Generalizing association rules. DAMI:02.
ƒ L. V. S. Lakshmanan, J. Pei, and J. Han. Quotient Cube: How to
Summarize the Semantics of a Data Cube. VLDB'02.
ƒ D. Xin, J. Han, X. Li, B. W. Wah. Star-Cubing: Computing Iceberg
Cubes by Top-Down and Bottom-Up Integration. VLDB'03.

126
Ref: FP for Classification and Clustering

ƒ G. Dong and J. Li. Efficient mining of emerging patterns:

Discovering trends and differences. KDD'99.
ƒ B. Liu, W. Hsu, Y. Ma. Integrating Classification and Association
Rule Mining. KDD’98.
ƒ W. Li, J. Han, and J. Pei. CMAR: Accurate and Efficient
Classification Based on Multiple Class-Association Rules. ICDM'01.
ƒ H. Wang, W. Wang, J. Yang, and P.S. Yu. Clustering by pattern
similarity in large data sets. SIGMOD’ 02.
ƒ J. Yang and W. Wang. CLUSEQ: efficient and effective sequence
clustering. ICDE’03.
ƒ X. Yin and J. Han. CPAR: Classification based on Predictive
Association Rules. SDM'03.
ƒ H. Cheng, X. Yan, J. Han, and C.-W. Hsu, Discriminative Frequent
Pattern Analysis for Effective Classification”, ICDE'07.

127

Ref: Stream and Privacy-Preserving FP Mining

ƒ A. Evfimievski, R. Srikant, R. Agrawal, J. Gehrke. Privacy Preserving

Mining of Association Rules. KDD’02.
ƒ J. Vaidya and C. Clifton. Privacy Preserving Association Rule Mining
in Vertically Partitioned Data. KDD’02.
ƒ G. Manku and R. Motwani. Approximate Frequency Counts over
Data Streams. VLDB’02.
ƒ Y. Chen, G. Dong, J. Han, B. W. Wah, and J. Wang. Multi-
Dimensional Regression Analysis of Time-Series Data Streams.
VLDB'02.
ƒ C. Giannella, J. Han, J. Pei, X. Yan and P. S. Yu. Mining Frequent
Patterns in Data Streams at Multiple Time Granularities, Next
Generation Data Mining:03.
ƒ A. Evfimievski, J. Gehrke, and R. Srikant. Limiting Privacy Breaches
in Privacy Preserving Data Mining. PODS’03.

128
Ref: Other Freq. Pattern Mining Applications

ƒ Y. Huhtala, J. Kärkkäinen, P. Porkka, H. Toivonen.

Efficient Discovery of Functional and Approximate
Dependencies Using Partitions. ICDE’98.
ƒ H. V. Jagadish, J. Madar, and R. Ng. Semantic
Compression and Pattern Extraction with Fascicles.
VLDB'99.
ƒ T. Dasu, T. Johnson, S. Muthukrishnan, and V.
Shkapenyuk. Mining Database Structure; or How to
Build a Data Quality Browser. SIGMOD'02.
ƒ K. Wang, S. Zhou, J. Han. Profit Mining: From Patterns
to Actions. EDBT’02.
129

Further Improvements of Mining Methods

ƒ AFOPT (Liu, et al. @ KDD’03)

ƒ A “push-right” method for mining condensed frequent pattern
(CFP) tree

ƒ Carpenter (Pan, et al. @ KDD’03)

ƒ Mine data sets with small rows but numerous columns
ƒ Construct a row-enumeration tree for efficient mining

ƒ FPgrowth+ (Grahne and Zhu, FIMI’03)

ƒ Efficiently Using Prefix-Trees in Mining Frequent Itemsets, Proc.
ICDM'03 Int. Workshop on Frequent Itemset Mining
Implementations (FIMI'03), Melbourne, FL, Nov. 2003

ƒ TD-Close (Liu, et al, SDM’06)

130
Assignments & Readings

ƒ Readings
» Chapter 5
ƒ Individual Project #1
» Ongoing

131

Next Session: Classification and Prediction

132