BIDE Efficient Mining of Frequent Closed SequencesJianyong Wang - PDF document

1 BIDE: Efficient Mining of Frequent Close
BIDE: Efficient Mining of Frequent Closed SequencesJianyong Wang and Jiawei Han University of Illinois at Urbana-ChampaignPresented by: Yi-Hung Wu Closed Frequent Sequence Mining Where w

2 ill data mining research go? Data Knowle
ill data mining research go? Data Knowledge Action Frequent Itemsets, Association Rules, Sequential Patterns, Clusters, Outliers… Text Classification, Web Usage Prediction, Feature Sele

3 ction,Anomaly Detection, … Microarray Cl
ction,Anomaly Detection, … Microarray Classification,“In Vivo”Spam Filtering,… Data Mining Multimedia Mining Web Mining Text Mining Stream Mining Biomedical, Financial, Geoscience, Telec

4 om, …Invisible (embeddedtool)Action-orie
om, …Invisible (embeddedtool)Action-oriented Applied Data MiningConstraint-based Closed Frequent Sequence Mining Where will data mining research go? Data Knowledge Action Frequent Itemse

5 ts, Association Rules, Sequential Patter
ts, Association Rules, Sequential Patterns, Clusters, Outliers… Text Classification, Web Usage Prediction, Feature Selection,Anomaly Detection, … Microarray Classification,“In Vivo”Spam

6 Filtering,… Data Mining Multimedia Mini
Filtering,… Data Mining Multimedia Mining Web Mining Text Mining Stream Mining Biomedical, Financial, Geoscience, Telecom, … Invisible (embeddedtool)Action-oriented Applied Data MiningC

7 onstraint-based Closed Frequent Sequence
onstraint-based Closed Frequent Sequence Mining Which patterns are interesting (applicable)? All Frequent Confidence Only Confident Optimal Interestingness Closed Frequent Sequence Minin

8 g Which patterns are interesting (applic
g Which patterns are interesting (applicable)? All Frequent Confidence Only Confident Optimal Interestingness Closed Frequent Top-k Frequent Circumstance Closed Frequent Sequence Mining

9 Which patterns are interesting (applicab
Which patterns are interesting (applicable)? All Frequent Confidence Only Confident Optimal Interestingness Closed Frequent Top-k Frequent Circumstance Aggregate Closed Frequent Sequence

10 Mining Which patterns are interesting (
Mining Which patterns are interesting (applicable)? All Frequent Confidence Only Confident Optimal Interestingness Closed Frequent Top-k Frequent Circumstance Aggregate Frequent Closed

11 Frequent Sequence Mining Take itemset as
Frequent Sequence Mining Take itemset as an example…–F Max F le T Closed Frequent Sequence Mining Take itemset as an example…–F Max F le T Closed Frequent Sequence Mining CHARM [Za

12 ki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X
ki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Frequent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–

13 Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c
Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Frequent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c

14 (Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y
(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Frequent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X

15 )=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise
)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Frequent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=

16 c(X4.Otherwise Closed Frequent Sequence
c(X4.Otherwise Closed Frequent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Freq

17 uent Sequence Mining CHARM [Zaki: sdm02,
uent Sequence Mining CHARM [Zaki: sdm02, kdd03]–IT-tree–Pruning! 1.t(X)=t(Y)c(X)=c(Y)=c(X2.t(X)c(Y), but c(X)=c(X3.t(X)c(Y), but c(Y)=c(X4.Otherwise Closed Frequent Sequence Mining –Lexi

18 cographic sequence tree–Pruning!Common
cographic sequence tree–Pruning!Common PrefixPartial OrderEarly Termination by Equivalence P.5 P.5 Closed Frequent Sequence Mining –Lexicographic sequence tree–Pruning!Common Prefix

19 Partial OrderEarly Termination by Equiv
Partial OrderEarly Termination by Equivalence P.5 P.5 Closed Frequent Sequence Mining –Lexicographic sequence tree–Pruning!Common PrefixPartial OrderEarly Termination by Equivalence

20 P.5 P.5 Closed Frequent Sequence Mining
P.5 P.5 Closed Frequent Sequence Mining –Lexicographic sequence tree–Pruning!Common PrefixPartial OrderEarly Termination by Equivalence P.5 P.5 Closed Frequent Sequence Mining How to

21 mine closed frequent sequences?–PrefixSp
mine closed frequent sequences?–PrefixSpan Prefix sequence lattice–Hashing: size, s-id sumSupport equalitySubsumption check Closed Frequent Sequence Mining How to mine closed frequent

22 sequences?–PrefixSpan Prefix sequence la
sequences?–PrefixSpan Prefix sequence lattice–Hashing: size, s-id sumSupport equalitySubsumption check P.6 Closed Frequent Sequence Mining How to mine closed frequent sequences?–Prefix

23 Span Prefix sequence lattice–Hashing: si
Span Prefix sequence lattice–Hashing: size, s-id sumSupport equalitySubsumption check P.6 P.6 Closed Frequent Sequence Mining How well does CloSpan perform?D10C10T2.5N10S6I2.5 Closed

24 Frequent Sequence Mining How well does C
Frequent Sequence Mining How well does CloSpan perform?D10C10T2.5N10S6I2.5 Closed Frequent Sequence Mining Can we mine closed frequent sequences withoutcandidate maintenance?–BI-Directi

25 onal ExtensionForward extension events
onal ExtensionForward extension eventsBackward extension events–Closure checkNo FENo BE–Pruning!BackScan extension eventsck Closed Frequent Sequence Mining Can we mine closed frequ

26 ent sequences withoutcandidate maintenan
ent sequences withoutcandidate maintenance?–BI-Directional ExtensionForward extension eventsBackward extension events–Closure checkNo FENo BE–Pruning!BackScan extension eventsck Cl

27 osed Frequent Sequence Mining Can we min
osed Frequent Sequence Mining Can we mine closed frequent sequences withoutcandidate maintenance?–BI-Directional ExtensionForward extension eventsBackward extension events–Closure chec

28 kNo FENo BE–Pruning!BackScan extensi
kNo FENo BE–Pruning!BackScan extension eventsck Closed Frequent Sequence Mining Where to find forward/backward extensions? , MP, MP, MP Closed Frequent Sequence Mining Where to find

29 forward/backward extensions?FE={locally
forward/backward extensions?FE={locally frequent items with full supports} , MP, MP, MP Closed Frequent Sequence Mining Where to find forward/backward extensions?FE={locally frequent ite

30 ms with full supports}For prefix ABC –La
ms with full supports}For prefix ABC –Last instance = C–LL: the i-th last-in-last appearanceLL–MP: the i-th maximum periodMP , MP, MP, MP Closed Frequent Sequence Mining Where to find

31 forward/backward extensions?FE={locally
forward/backward extensions?FE={locally frequent items with full supports}For prefix ABC –Last instance = C–LL: the i-th last-in-last appearanceLL–MP: the i-th maximum periodMP–Scan ba

32 ckward each of MPScanSkip , MP, MP, MP C
ckward each of MPScanSkip , MP, MP, MP Closed Frequent Sequence Mining How does BIDE improve the mining efficiency? Closed Frequent Sequence Mining How does BIDE improve the mining effic

33 iency?BackScan: ABC –LFLF–SMP: the i-th
iency?BackScan: ABC –LFLF–SMP: the i-th semi-maximum periodSMP Closed Frequent Sequence Mining How does BIDE improve the mining efficiency?BackScan: ABC –LFLF–SMP: the i-th semi-maxim

34 um periodSMPi,e appears in each of SMP–
um periodSMPi,e appears in each of SMP–Stop projection! Closed Frequent Sequence Mining How does BIDE improve the mining efficiency?BackScan: ABC –LFLF–SMP: the i-th semi-maximum perio

35 dSMPi,e appears in each of SMP–Stop pro
dSMPi,e appears in each of SMP–Stop projection! Closed Frequent Sequence Mining How does BIDE improve the mining efficiency?BackScan: ABC –LFLF–SMP: the i-th semi-maximum periodSMPi,e

36 appears in each of SMP–Stop projection!
appears in each of SMP–Stop projection! Closed Frequent Sequence Mining Does BIDE perform much better? PrefixSpan/SPADE when support threshold is lowBIDE consumes much less memoryand ca

37 n be an order of magnitude fasterthan Cl
n be an order of magnitude fasterthan CloSpanBIDE has linear scalability in terms of data sizeeffective in enhancing the performance Closed Frequent Sequence Mining Does BIDE perform muc

38 h better? PrefixSpan/SPADE when support
h better? PrefixSpan/SPADE when support threshold is lowBIDE consumes much less memoryand can be an order of magnitude fasterthan CloSpanBIDE has linear scalability in terms of data size

39 effective in enhancing the performance C
effective in enhancing the performance Closed Frequent Sequence Mining Does BIDE perform much better? PrefixSpan/SPADE when support threshold is lowBIDE consumes much less memoryand can

40 be an order of magnitude fasterthan CloS
be an order of magnitude fasterthan CloSpanBIDE has linear scalability in terms of data sizeeffective in enhancing the performance Closed Frequent Sequence Mining Does BIDE perform much

41 better? PrefixSpan/SPADE when support th
better? PrefixSpan/SPADE when support threshold is lowBIDE consumes much less memoryand can be an order of magnitude fasterthan CloSpanBIDE has linear scalability in terms of data sizeef

42 fective in enhancing the performance Clo
fective in enhancing the performance Closed Frequent Sequence Mining Conclusion RemarksClosed Frequenthas the same expressive power as All Frequent, but provides more compactresults and

43 likely betterefficiency.projectionpatter
likely betterefficiency.projectionpattern-a new paradigm without candidate maintenance Closed Frequent Sequence Mining Any Question? 3| ()(ີ.;䀀)(a)(bd) Closed Frequent Sequence Mi

44 ning Any Question? My Questions… 3| ()(&
ning Any Question? My Questions… 3| ()(ີ.;䀀)(a)(bd) Closed Frequent Sequence Mining Any Question? –.80;(e)(b) vs. ()()()(&#x-7.2;w)(y)(z)(&#x-6.5;w)(x)(y)()(x&#xz-6.;瀀)(y

45 ) 3| ()(ີ.;䀀)(a)(bd) Closed Frequ
) 3| ()(ີ.;䀀)(a)(bd) Closed Frequent Sequence Mining Any Question? –.80;(e)(b) vs. ()()()(&#x-7.2;w)(y)(z)(&#x-6.5;w)(x)(y)()(x&#xz-6.;瀀)(y)How to efficiently compute or

46 maintain MPDoes it easily adapt BIDE to
maintain MPDoes it easily adapt BIDE to sequences of itemsets? 3| ()(ີ.;䀀)(a)(bd) Closed Frequent Sequence Mining Any Question? –.80;(e)(b) vs. ()()()(&#x-7.2;w)(y)(z)(&#x-

47 6.5;w)(x)(y)()(x&#xz-6.;瀀)(y)How to ef
6.5;w)(x)(y)()(x&#xz-6.;瀀)(y)How to efficiently compute or maintain MPDoes it easily adapt BIDE to sequences of itemsets?What is the difference between Closed Frequent 3| ()(ີ.;ä