1 Global-as-View and Local-as-View for Information Integration CS652 Spring 2004 Presenter: Yihong Ding.

1

Global-as-View and Local-as-Viewfor Information Integration

CS652 Spring 2004

Presenter: Yihong Ding

2

Common Integration Architecture

• Information Integration Systems

• Global-as-view (Gav.) vs. Local-as-view (Lav.)

• Query Reformulation• Specification of Source

Description• Adding new sources

3

Query Reformulation

• Problem: rewrite a user query expressed in the mediated schema into a query expressed in the source schema

Given a query Q in terms of the mediator schema relations, and descriptions of information sources

Find a query Q’ that uses only the source relations, such that– Q’ Q, and– Q’ provides all possible answers to Q given the sources

4

Solving Queries by Views

Mediator Relations

Source Relations

5

Query Rewriting Using Views

• Query Containment: q’ q D q’(D) q(D)• Query Equivalence: q’=q q’ q ^ q q’Given query q and view definitions V={v1, …, vn}• q’ is an Equivalent Rewriting of q using V if

– q’ refers only to views in V, and– q’ = q

• q’ is an Maximally-Contained Rewriting of q using V if – q’ refers only to views in V and– q’ q, and– There is no rewriting q1, such that q’ q1 and q1q’

6

ComputationComplexity

p

k

p

k

pk

p

k

p

k 1

7

Complexity of Query Containment

• Conjunctive Queries (CQ) (NP-Complete) – Q1: p(X,Z) :- a(X,Y) & a(Y,Z)– Q2: p(X,Z) :- a(X,Y) & a(V,Z)

• CQ’s With Negation ( -Complete)– Q1: p(X,Z) :- a(X,Y) & a(Y,Z) & NOT a(X,Z)

• CQ’s With Arithmetic Comparison ( -Complete)– Q1: p(X,Z) :- a(X,Y) & a(Y,Z) & X<Y

• Datalog Programs– p(A,C) :- a(A,B) & b(B,C)

p

2

p

2

8

Specification of Source Description

• Views: resources that used by integrator to help to answer queries

• Gav. Mediator relation defined as view over source relations

• Lav. Source relation defined as view over mediator relations

9

Information Integration Systems

• Tsimmis– Stanford and IBM– Global-as-View (Gav)– Mediator relations defined as views of source relations

• Information Manifold (IM)– AT&T– Local-as-View (Lav)– Description logic– Source relations defined as views of mediator relations ( a

collection of global predictions)

10

TSIMMIS – Gav Solution

• The Stanford-IBM Manager of Multiple Information Sources (TSIMMIS)

• Offers:– A flexible data model– A common query language– Other supporting tools

11

TSIMMIS – Components

• OEM (Object-Exchange Model)

• LOREL (Lightweight Object REpository Language)

• MSL (Mediator Specification Language)

• Wrappers

12

TSIMMIS – OEM

• Object Exchange Model• The data model for TSIMMIS• “self-describing” (labels carry all of the

information that there is about an object)• Flexible• First order logic

13

TSIMMIS – OEM

OID: label type value

Object Identifier

Human Understandable

“set” or “string”

A set or a string

14

TSIMMIS – OEM

library

set

book set

author

string

title string

Aho

Compilers…

15

TSIMMIS – OEMFirst order predicate logic

author

string

Aho123

author( T, “Aho” )

This would return the object IDs of allobjects with a label “author” and value “Aho”.

16

TSIMMIS – LOREL

• Lightweight Object REpository Language• An OQL for OEM• The end-user language for TSIMMIS

17

TSIMMIS – LOREL

• Example

select library.book.titlefrom librarywhere library.book.author = “Aho”

18

TSIMMIS – LOREL

• Partial Match Semantics

select R.Afrom R, S, Twhere R.A = S.A or R.A = T.A

• This would fail to return anything in SQL if either S or T were empty.

• Because of partial match semantics this does not fail in LOREL

19

TSIMMIS – MSL

• Mediator Specification Language• Allows declarative specification of mediators• Object oriented, logical query language• Targeted to OEM

20

TSIMMIS – MSL Query

Mediator

Mediator

WrapperWrapper

SourceSource

<booktitle X> :- <library { <book { <title X> <author “Aho”> } > } > @s1

library

set

book set

author

string

Aho

title string

Compilers…

21

TSIMMIS – Wrappers Query

Mediator

Mediator

WrapperWrapper

SourceSource

• Wrappers are similar to database drivers

• Wrappers are written with MSL

22

TSIMMIS – Wrappers

• Wrappers have the form:

MSL template// action //

• Example:

<books X> :- <library { X:<book {<title X> <author $AU>}> }>@s1// sprintf(lookup-query, “find author %s”, $AU) //

23

TSIMMIS – Summary

• End users need to specify their sources w.r.t. a mediator model – OEM in TSIMMIS

• Query specification is standard – LOREL • Query rewriting is straightforward – MSL and

wrappers • To add a new source is not easy – need to

specify it in the mediator model

24

Information Manifold

• Challenges for Information Integration– Interrelated data over

multiple information sources

– Large number of the sources

– Limited size of data in many of the sources

– Greatly variant details of interacting with each source

25

IM Architecture

1

2 3

Bucket algorithm

26

World View

Product(Model)Automobile(Model, Year, Category)Motorcycle(Model, Year)Car(Model, Year, Category) NewCar(Model, Year, Category)UsedCar(Model, Year, Category) CarForSale(Model, Year, Category, Price, SellerContact)

Automobile

Car Motorcycle

Car

UsedCar CarForSale

Product

Automobile

Virtual Relations:

Classes:

NewCar

27

Source Descriptions

For each source:

• Content Record • Capability Record

Web Sources forAutomobile Application

28

Content Records of Auto Sources

29

Capability Records of Auto Sources

desired input set possible output set

capable selection set

30

Query Reformulation

• Containing instead of equivalent– Incomplete source – Useful subset

• Utilizes Plan Generator to:– Prune irrelevant sources– Split query into subgoals– Generate conjunctive query plans– Find executable ordering of subgoals

31

The Bucket Algorithm

Given: user query q, source descriptions {Vi}

1. Find relevant source (fill buckets) For each relation g in query q

• Find Vj that contains relation g

• Check that constraints in Vj are compatible with q

2. Combine source relations {Vj} from each bucket into a conjunctive query q’ and check for containment (q’ q)

32

The Bucket Algorithm: Example

q(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992, Model(c,m), Price(c,p), ProductReview(m,y,r)

q(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992, Model(c,m), Price(c,p), ProductReview(m,y,r)

33

1. Filling the Buckets

q(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992,

Model(c,m), Price(c,p), ProductReview(m,y,r)

q(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992,

Model(c,m), Price(c,p), ProductReview(m,y,r)

V1(c1)

V2(c2)

V3(c3)

V1(c1,t1)

V2(c2,t2)

V3(c3,t3)

V1(c1,y1)

V2(c2,y2)

V3(c3,y3)

V1(c1,m1)

V2(c2,m2)

V3(c3,m3)

V1(c1,p1)

V2(c2,p2)

V3(c3,p3)

V5(m5,y5,r5)

CarForSale(c), Category(c,t), Year(c,y), Model(c,m), Price(c,p), ProductReview(m,y,r)y1992t=sportscar

34

2. Checking Containment

User Queryq(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992,

Model(c,m), Price(c,p), ProductReview(m,y,r)

User Queryq(m,p,r) CarForSale(c), Category(c,sportscar), Year(c,y), y1992,

Model(c,m), Price(c,p), ProductReview(m,y,r)

Result Queryq’(m,p,r) V1(c)({Category(c):sportscar}, {Price(c), Model (c), Year(c)}, {Year(c)1992, Category(c)=sportscar}), V5(m,y,r)({m:Model(c), y:Year(c)}, {r}, {}).

Result Queryq’(m,p,r) V1(c)({Category(c):sportscar}, {Price(c), Model (c), Year(c)}, {Year(c)1992, Category(c)=sportscar}), V5(m,y,r)({m:Model(c), y:Year(c)}, {r}, {}).

?

Expanded Queryq’(m,p,r) CarForSale(c), UsedCar(c), Category(c,t), t=sportscar, Model(c,m), Year(c,y), Price(c,p), ProductReview(m,y,r), y1992

Expanded Queryq’(m,p,r) CarForSale(c), UsedCar(c), Category(c,t), t=sportscar, Model(c,m), Year(c,y), Price(c,p), ProductReview(m,y,r), y1992

35

Finding an Executable Ordering

CarForSale(c), Category(c,t), Year(c,y), Model(c,m), Price(c,p), ProductReview(m,y,r)y1992t=sportscar

V1(c) V1(c,t) V1(c,y) V1(c,m) V1(c,p) V5(m,y,r)

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s)}

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s)}

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s), ProductReview(m,y,r)}

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s), ProductReview(m,y,r)}

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s), ProductReview(m,y,r), y1992}

BindAvail1 = {CarForSale(c,sportscar), Model(c,m), Year(c,y), Price(c,p), SellerContact(c,s), ProductReview(m,y,r), y1992}

36

Advantages and Disadvantages

• Gav: Tsimmis– Advantage

• Query reformulation: rule unfolding– Disadvantage

• Mediation description• Adding, removing, and modifying source description

– Better for static, centralized systems

• Lav: Information Maniford– Advantage: adding new sources

• Mediator (global predicates, source descriptions)• Query processing

– Disadvantages• query reformulation (Bucket algorithm)

– Better for dynamic, distributed systems