Creating Probabilistic Databases from InformationExtraction Models

Rahul GuptaIBM Research LabNew Delhi, India

rahulgupta@in.ibm.com

Sunita SarawagiIndian Institute of Technology

Bombay, India

sunita@iitb.ac.in

ABSTRACT

1. INTRODUCTION

Permission to copy without fee all or part of this material is granted providedthat the copies are not made or distributed for direct commercial advantage,the VLDB copyright notice and the title of the publication and its date appear,and notice is given that copying is by permission of the Very Large DataBase Endowment. To copy otherwise, or to republish, to post on serversor to redistribute to lists, requires a fee and/or special permission from thepublisher, ACM.VLDB ‘06, September 12-15, 2006, Seoul, Korea.Copyright 2006 VLDB Endowment, ACM 1-59593-385-9/06/09

965

Outline

2. SETUP

2.1 Models for automatic extraction

X

P P

X X

966

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Pre

cisi

on o

f top

seg

men

tatio

n

Probability of top segmentation

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Pre

cisi

on o

f top

seg

men

tatio

n

Probability of top segmentation

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

>20050-200

41-5031-40

21-3011-20

4-10321

Rel

ativ

e F

requ

ency

Number of segmentations required

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

>10050-100

41-5031-40

21-3011-20

4-10321

Rel

ativ

e F

requ

ency

Number of segmentations required

X

P

8><>:

2.2 Database representation of extraction un-certainty

2.2.1 Segmentation per row model

X

967

2.2.2 One-row model

Y

2.2.3 Multi-row model

X Y

2.3 Quantifying approximation quality

968

X

3. APPROXIMATIONS

3.1 One row model

X

X X

X X

X

3.1.1 Computing marginals from model

X

X X

P P

X X

P

X X

X

3.2 Multi-row model

969

X X

Q

3.3 Enumeration-based approach

X X

P

PP

X

s2

s 3

s s1 4,

A

B’West’,_

’52−A’,House_no

3.4 Structural approach

X X

X X

X X

970

X

P

P

3.5 Merging structures

PP

X

971

P

PP

P

P

P P

4. EXPERIMENTS

4.1 Datasets

4.2 Error reduction with probabilistic data

972

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6

Avg

KL

dive

rgen

ce

Number of rows

Multi-rowSegmentation-per-row

00.5

11.5

22.5

33.5

44.5

5

0 1 2 3 4 5 6

Avg

KL

dive

rgen

ce

Number of rows

Multi-rowSegmentation-per-row

4.3 Comparing models of imprecision

4.4 Evaluating the multi-row algorithm

Dependence on input parameters

973

0

20

40

60

80

100

120

140

160

0 10 20 30 40 50 60 70 80

k re

qd to

mat

ch S

truc

tMer

ge

No. of initial partitions chosen till 0.05 cutoff

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.02 0.04 0.06 0.08 0.1

Ave

rage

div

erge

nce

Info gain cutoff

Address (strong)Address (weak)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0 0.02 0.04 0.06 0.08 0.1

Ave

rage

div

erge

nce

Info gain cutoff

Cora (strong)Cora (weak)

Impact on query results

5. RELATED WORK

974

0

0.05

0.1

0.15

0.2

0.25

-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Fra

ctio

n of

str

ings

Ranking Inversions

OneRowStruct+Merge

0

0.5

1

1.5

2

2.5

0 0.1 0.2 0.3 0.4 0.5 0.6

Ran

king

Inve

rsio

ns

Divergence

OneRow

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0 0.1 0.2 0.3 0.4 0.5 0.6

Ran

king

Inve

rsio

ns

Divergence

StructMerge

6. CONCLUSIONS

Acknowledgements

7. REFERENCES

975

0

0.5

1

1.5

2

2.5

0 50 100 150 200 250 300 350 400 450

Div

erge

nce

String Id

StructuralOne RowStruct+MergeEnumeration

0

0.5

1

1.5

2

2.5

3

0 20 40 60 80 100 120 140

Div

erge

nce

String Id

StructuralOne RowStruct+MergeEnumeration

976