Università di AnconaCoopIS01 September 6, 20011 C OOPERATION S TRATEGIES FOR I NFORMATION I NTEGRATION Maurizio Panti, Luca Spalazzi, Loris Penserini {panti,spalazzi,pense}@inform.unian.it.

September 6, 2001 1

Università di AnconaCoopIS’01

COOPERATION STRATEGIES FOR

INFORMATION INTEGRATION

Maurizio Panti, Luca Spalazzi, Loris Penserini

{panti,spalazzi,pense}@inform.unian.it

Istituto di Informatica - Università degli studi di Ancona

September 6, 2001 2


Talk Overview

Motivations and Goals Local strategies Cooperation strategies

the choice of partnersthe choice of queriesthe choice of answers

Discussion

September 6, 2001 3


Motivation

Information systems are collections of information sources and information consumers: Distributed Heterogeneous

at physical levelat logical levelat conceptual level (names and schemas)

Dynamic changes of information sources or their schemaschanges of information consumers or their needs

September 6, 2001 4


Goal

Rewriting a consumer’s query into queries to specific information sources

when we have

a distributed, heterogeneous, and strongly dynamic information system.

September 6, 2001 5


Usually query rewriting and information integration systems adopt the

Mediator Architecture:

[TSIMMIS, Squirrel, WHIPS, Carnot, SIMS, Information Manifold, Infomaster]

dynamic sources: systems are overloaded with expensive updating operations;

dynamic consumers: systems do not perform user profiling.

Related Work

consumer consumer

mediator mediator

... ...

source sourcesource ...

consumer

September 6, 2001 6


consumer consumer

mediator mediator

... ...


consumersource: w1

publication

journal conference

acm_trans ieee_trans

... ...

...people =

pub.publication name.String affiliation.String

publication =title.String booktitle.String publisher.String keyword.String

Wrapper

DBMS

publication

titlebooktitlepublisherkeyword

people

pub name affiliation

wrappera description logic as data modelling

and query language [e.g. C-Classic]source query processing is based on rewriting query using views over the local source [adapted by Beeri, Levy, Rousset]

Information Source

September 6, 2001 7


consumer consumer

mediator mediator

... ...


consumer

Mediator

mediator

Thesaurus

Mediated Schema

mediated schemaa description logic as data modelling

and query language [e.g. C-Classic]query processing is based on rewriting query using views over the distributed sources [adapted by Beeri, Levy, Rousset]

September 6, 2001 8


Retrieval: Conjunction of concepts that are maximally contained in the query

Rewriting: Composition of rewriting of retrieved concepts

query: Q = pub.(ai ∏ db) ∏ pub.acm

Rewriting query using views

T

journal

type.{“Trans”}

acm_trans

acm ai

agents

db pub.(journal db)

J=pub.ai

K=pub.(agents db)

I=pub..acm_journalacm_journal

K=pub.(agents

db)

pub.(ai db)

H J = pub.(ai db journal)

I=pub..acm_journal

pub.acm

retrieved concepts: {K, (H ∏ J)} { I }rewrite: (view(K)∏view(I))U(view(H)∏view(J)∏view(I))

September 6, 2001 9


source: w1

conference

publication

journal

acm_trans ieee_trans... ...

...

acm_trans =journal

type.{“Trans”}

publisher.{“ACM”}

.

.

...

journal =title.String

booktitle.String

type.{“Trans”,”Magazine”}

publisher.String

keyword.String

source: w3

people = pub.ai

name.String

position.String

.

.

ai

planning cbr agents

... ... ...

...

ai =title.String

booktitle.String


publisher.String

year.Integer

source: w2

db =title.String

booktitle.String

type.String

year.Integer

.

people =pub.db

name.String

degree.String

.

db

object-oriented active federated

... ... ...

...

#id pub… …..

#id pub… …..

Mediator M

rewrite: (view(K)∏view(I))U(view(H)∏view(J)∏view(I))

Query Execution

#id pub… …..

rewrite: (view(K)∏view(I))U(view(H)∏view(J)∏view(I))data: #id pub

… …..… …..… …..

September 6, 2001 10


Local Failures

In query rewriting:the mediator is not able to rewrite (some or all the components) of the input query.

In query execution:the mediator is not able to execute (some or all the components) of the rewrited query.



Cooperation Strategies

consumer consumer

mediator mediator

... ...


consumer

mediator

Thesaurus

Mediated Schema

all the mediators

partner

mediator

source

new mediators

succeeding mediators

failing mediators

all the sources

new sources

succeeding sources

failing sources

originalwhole

single component

selectedconcepts

whole

single component

rewritingwhole

single component

queryanswer

rewriting

data



Cooperation with MediatorsAsking for Rewriting

Mediator MMediator M Mediator NMediator N

W1W1 W2W2 W3W3

1: request

3: query4: data

2: rewrite




W1W1 W2W2 W3W3

1: request

2: query 3: data

4: data

Cooperation with MediatorsAsking for Data



Cooperation with Sources


W1W1 W2W2 W3W3

2: rewrite3: data

1: request



Strategy Comparison

Partner Type of Efficiency Effectiveness Answer No. of Messages Redund. Recall Prec. 1st. time Next time

Source (Sj) Rewriting 4s 2s No 1 1 “ Data 2s 2s No 1 1

Mediator (Ni) Rewriting 2m+2s’ 2s’ Yes 1 No “ Data 2m+2ms’ 2m+2ms’ No 1 No

m: number of mediators

s: number of sources

s’: number of sources that cooperate with Ni



Strategy Comparison

Query RewritingOriginal Query Q Sol N (Q)Selected Concepts Q’ Sol N (Q’)Rewriting of selected concepts view(Q’) Sol N (view(Q’))

SolN(view(Q’)) SolN(Q’) SolN(Q)



Conclusion

1st scenario (s ≈ m+s’)

mediators can be used for user profiling, mediators can be used to solve name

heterogeneity and integrate data, in order to solve schema heterogeneity, for a

mediator the most efficient and effective strategy is to directly cooperate with sources,

in order to update its schemas, for a mediator a “lazy” approach can be not appropriate.



Conclusion

2nd scenario (s >> m+s’)

mediators can be used for user profiling, the most efficient strategy is the cooperation

with other mediators, cooperation with wrappers is useful only

when mediators are not able to rewrite a given query,

in order to update its schemas, for a mediator a “lazy” approach is appropriate.





M’s Mediated Schema

Problem Solution (Query) Rewriting Information Sources

pub. (journalΠdb)

(pub.journal Π pub.db), (pub.ai Π pub.db)

(pub.journal, w1), (pub.db,w2), (pub.ai,w3)

I=pub. acm_journal

(pub.acm_trans), (pub.type.{“Trans”} Π pub.publisher.{“ACM”})

(pub.acm_trans, w1), (pub.type.{“Trans”}, w3), (pub.publisher.{“ACM”}, w3)

J=pub.ai (pub.keyword.{“AI”}), (pub.ai)

(pub.keyword.{“AI”}, w1), (pub.ai, w3)

K=pub. (agentsΠdb)

(pub.keyword.{“Agents”} Π pub.db), (pub.agents Π pub.db)

(pub.keyword.{“Agents”}, w1) (pub.db, w2), (pub.agents, w3 )




Composition of rewriting of retrieved concepts

view( pub.(ai ∏ db) ∏ pub.acm ) =

= view( pub.(ai ∏ db) ) ∏ view( pub.acm )

= (view(K)∏view(I))U(view(H)∏view(J)∏view(I))

= … ...




={ (pub. keyword.{“Agents”} ∏ pub.db ∏ pub.acm_trans) ,

(pub.keyword.{“Agents”}∏pub.db ∏ pub.type.{“Trans”}∏pub.publisher.{“ACM”}),

(pub.agents ∏ pub.db ∏ pub.acm_trans) ,

(pub.agents ∏ pub.db ∏ pub.type.{“Trans”}∏pub.publisher.{“ACM”}),

(pub.journal ∏ pub.db ∏ pub. keyword.{“AI”} ∏ pub.acm_trans) ,

…}

=(view(K) ∏ view(I)) U (view(H) ∏ view(J) ∏ view(I))



Local Failure in Query Rewriting

query: Q=pub.(ai∏db)∏affiliation.{“Stanford”}

T

journal

type.{“Trans”}

acm_trans

acm ai

agents

db pub.(journal db)

J=pub.ai

K=pub.(agents db)

I=pub..acm_journalacm_journal

affiliation.{“Stanford”} K=pub.(agents

db)

pub.(ai db)

H J = pub.(ai db journal)

retrieved concepts: {K, (H ∏ J)} Ørewrite: failure



source: w1

conference

publication

journal


...

acm_trans =journal

type.{“Trans”}


.

.

...


booktitle.String


publisher.String

keyword.String

source: w3

people = pub.ai

name.String

position.String

.

.

ai

planning cbr agents

... ... ...

...

ai =title.String

booktitle.String


publisher.String

year.Integer

source: w2

db =title.String

booktitle.String

type.String

year.Integer

.

people =pub.db

name.String

degree.String

.

db


... ... ...

...

Local Failure in Query Execution

no a

nsw

er

#id pub… …..

#id pub… …..

Mediator M

rewrite: (view(K)∏view(I))U(view(H)∏view(J)∏view(I))rewrite: (view(K)∏view(I))U(view(H)∏view(J)∏view(I))data: failure



T

acm_tods

acm

pub.acm_tods

pub.acm_tocl

acm_tocl

Mediator N

query: pub.acm

Mediator M


rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}

pub.acm_tods

pub.acm_tocl

pub.acm

pub.acm_tods

pub.acm_tocl

retrieved concepts:{pub.acm_tocl, pub.acm_tods}



Mediator M


rewrite:{view(pub.acm_tocl),view(pub.acm_tods)} rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}data: #id pub

… …..… …..… …..

source: w1

conference

publication

journal


...

acm_trans =journal

type.{“Trans”}


.

.

...


booktitle.String


publisher.String

keyword.String

source: w3

people = pub.ai

name.String

position.String

.

.

ai

planning cbr agents

... ... ...

...

ai =title.String

booktitle.String


publisher.String

year.Integer

source: w2

db =title.String

booktitle.String

type.String

year.Integer

.

people =pub.db

name.String

degree.String

.

db


... ... ...

...



T

acm_tods

acm

pub.acm_tods

pub.acm_tocl

acm_tocl

Mediator N

query: pub.acm

Mediator M

pub.acm_tods

pub.acm_tocl

pub.acm

pub.acm_tods

pub.acm_tocl


retrieved concepts:{pub.acm_tocl, pub.acm_tods} rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}



T

acm_tods

acm

pub.acm_tods

pub.acm_tocl

acm_tocl

Mediator N

pub.acm_tods

pub.acm_tocl

pub.acm

pub.acm_tods

pub.acm_tocl


rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}

source: w1

conference

publication

journal


...

acm_trans =journal

type.{“Trans”}


.

.

...


booktitle.String


publisher.String

keyword.String

source: w3

people = pub.ai

name.String

position.String

.

.

ai

planning cbr agents

... ... ...

...

ai =title.String

booktitle.String


publisher.String

year.Integer

source: w2

db =title.String

booktitle.String

type.String

year.Integer

.

people =pub.db

name.String

degree.String

.

db


... ... ...

...

rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}data: #id pub

… …..… …..… …..



T

acm_tods

acm

pub.acm_tods

pub.acm_tocl

acm_tocl

Mediator N

pub.acm_tods

pub.acm_tocl

pub.acm

pub.acm_tods

pub.acm_tocl


rewrite:{view(pub.acm_tocl),view(pub.acm_tods)} rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}data: #id pub

… …..… …..… …..

query: pub.acm

Mediator M

rewrite:{view(pub.acm_tocl),view(pub.acm_tods)}data: #id pub

… …..… …..… …..



Cooperation with Sourcessource: w1

conference

publication

journal


...

acm_trans =journal

type.{“Trans”}


.

.

...


booktitle.String


publisher.String

keyword.String

source: w3

people = pub.ai

name.String

position.String

.

.

ai

planning cbr agents

... ... ...

...

ai =title.String

booktitle.String


publisher.String

year.Integer

source: w2

db =title.String

booktitle.String

type.String

year.Integer

.

people =pub.db

name.String

degree.String

.

db


... ... ...

...

source: w4

....

acm_tocl = acm_trans

booktitle.{“TOCL”}

acm_tods = acm acm_transbooktitle.{“TODS”}

.acm_jacm =acm journal

booktitle.{“JACM”}

.

acm_ccs =acm

booktitle.{“CCS”}

type.{“Proc”}

.

conference journal

acm_transacm_jacm

.........

acm_ccs

acm_tods acm_tocl

acm

query: pub.acm

Mediator M

source: w4

....

acm_tocl = acm_trans

booktitle.{“TOCL”}

acm_tods = acm acm_transbooktitle.{“TODS”}

.acm_jacm =acm journal

booktitle.{“JACM”}

.

acm_ccs =acm

booktitle.{“CCS”}

type.{“Proc”}

.

conference journal

acm_transacm_jacm

.........

acm_ccs

acm_tods acm_tocl

acm

retrieved:{pub.acm_ccs,pub.acm_jacm,pub.acm_tods}

rewrite:{view(pub.acm_ccs),view(pub.acm_jacm), view(pub.acm_tods)}

data:

Mediator M

#id pub… …..… …..… …..



Redundancy

Cn(M) mediated schema of M after n interactions with N

C(N) mediated schema of N

))((

))()((

NCcard

NCMCncard



Recall

Cn(M) mediated schema of M after n interactions with S1, … Sn

information need of a consumer(a view of S1, … Sn )

)(

))((

card

MCncard



Precision



))((

))((

MCncard

MCncard



N’s Mediated Schema


pub. acm_tods

(pub.booktitle.{“TODS”} Π pub.acm)

(pub.booktitle.{“TODS”}, w4), (pub.acm , w4)

pub. acm_tocl

(pub.booktitle.{“TOCL”} Π pub.acm_trans)

(pub.booktitle.{“TODS”}, w4), (pub.acm_trans , w1)



M’s Mediated Schema (updated)


H (pub.journal Π pub.db), (pub.ai Π pub.db)


I (pub.acm_trans), (pub.type.{“Trans”} Π pub.publisher.{“ACM”})


J (pub.keyword.{“AI”}), (pub.ai)


K (pub.keyword.{“Agents”} Π pub.db), (pub.agents Π pub.db)


pub.acm (pub.booktitle.{“TODS”} Π pub.acm)

(pub.booktitle.{“TODS”}, w4), (pub.acm , w4)













pub.acm (pub.acm_tods) (pub.acm_tods , N)













pub.acm (pub.acm_ccs) , (pub.acm_jacm) , (pub.acm_tods)

(pub.acm_ccs , w4) , (pub.acm_jacm , w4) , (pub.acm_tods , w4)



Cooperation with Mediators

Theorem.


C(N) mediated schema of N

limn

card (Cn(M)C(N))

card (C(N))1redundancy:



Cooperation with Mediators

Theorem.



limn

card (Cn(M) )

card ( )1recall:

limn

card (Cn(M) )

card(Cn(M)) indeterminateprecision:



Cooperation with Sources

Theorem.



recall:

precision:

limn

card (Cn(M) )

card ( )1

limn

card (Cn(M) )

card(Cn(M))1

Università di AnconaCoopIS01 September 6, 20011 C OOPERATION S TRATEGIES FOR I NFORMATION I NTEGRATION Maurizio Panti, Luca Spalazzi, Loris Penserini {panti,spalazzi,pense}@inform.unian.it.

Documents

rewriting query

query rewriting

data slide

sources mediator

rewriting mediator

rewrite slide

consumers query

request slide