The MOMIS approach for Information Integration

36
Domenico Beneventano 1 DB Group @ unimo The MOMIS approach for The MOMIS approach for Information Integration Information Integration Domenico Beneventano Domenico Beneventano

description

The MOMIS approach for Information Integration. Domenico Beneventano. Integration : motivation and problems. Internet has as side effect: the “fragmentation” of information in data sources of small size - PowerPoint PPT Presentation

Transcript of The MOMIS approach for Information Integration

Page 1: The MOMIS approach for  Information Integration

Domenico Beneventano1

DB

Gro

up @

uni

mo

The MOMIS approach for The MOMIS approach for Information IntegrationInformation Integration

Domenico BeneventanoDomenico Beneventano

Page 2: The MOMIS approach for  Information Integration

Domenico Beneventano2

DB

Gro

up @

uni

mo

Integration : motivation and problemsIntegration : motivation and problems

• Internet has as side effect: the “fragmentation” of information in data sources of small size

• Companies ask for an increased capability of analyzing data from heterogeneous data sources coming from legacy systems

• Efficient Network technologies make it convenient to develop distributed information systems

• An integration process has to take into account the problems deriving from the heterogeneity of data sources :

– Platforms Heterogeneity, it involves hardware/software/DBMSs Solution: different standards, as SQL, ODBC, XML, …

– Structural Heterogeneity, it involves the different conceptual, logic data models and the different representations of the same concept

– Semantic Heterogeneity, it involves the different meaning of terms in data sources (homonymy and synomymy)

Page 3: The MOMIS approach for  Information Integration

Domenico Beneventano3

DB

Gro

up @

uni

mo

Approaches for interoperability and Approaches for interoperability and integrationintegration

• Different approaches have been proposed to guarantee interoperability and integration of data sources:

– Read-only integrated Views : Mediation. The goal of the Mediator based architectures is to obtain an integrated view, read only, of the data stored in different sources.

Essentially, it is necessary to build a component, the Mediator, able to build an integrated schema and to allow users to pose queries with respect to this schema, receive a unified answer, thus disregarding the heterogeneity of the different sources.

– Databases which share information: Federation. This architecture implement a framework that allows database to behave as in a federation. Each source may extend its schema by including subschemas (and data) of other federated databases (Information Integrator IBM).

Page 4: The MOMIS approach for  Information Integration

Domenico Beneventano4

DB

Gro

up @

uni

mo

DB1

User MEDIATOR

Wrapper Wrapper Wrapper

...FileSystem

Wrapper/Mediator Architecture Wrapper/Mediator Architecture

Wrapper extracting the schema of the data sources : local schema

(there is one wrapper for each different kind of data source).

converting the local schema into the common format of the Mediator

sending a query to the data sources exporting the local queries results to the Mediator.

Page 5: The MOMIS approach for  Information Integration

Domenico Beneventano5

DB

Gro

up @

uni

mo

Semantic Integration of Heterogeneous Semantic Integration of Heterogeneous DataData

Global Virtual View(GVV)

The Local Schema of each source is available (source wrapping)

Data integration provides a Global Virtual View (GVV) that is a conceptualization describing the involved sources. allows a user to raise a query and to receive a single

unified answer

Source Source Source

LocalSchema

LocalSchema

LocalSchema

Mapping

Query

Page 6: The MOMIS approach for  Information Integration

Domenico Beneventano6

DB

Gro

up @

uni

mo

Main problems in data integration Main problems in data integration [Lenzerini 2003][Lenzerini 2003]

(Automatic) source wrapping

How to construct the Global Virtual View

How to discover interschema properties among the sources

and mappings between the sources and the Global Virtual

View

How to model the mappings between the sources and the

Global Virtual View

How to process updates expressed on the Global Virtual

View, and updates expressed on the sources (Schema

Evolution)

How to answer queries expressed on the Global Virtual

View (Global Query Management)

Query optimization

Data extraction, cleaning and reconciliation (Extensional

Integration)

Page 7: The MOMIS approach for  Information Integration

Domenico Beneventano7

DB

Gro

up @

uni

mo

MOMIS (Mediator envirOnment for Multiple Information Sources) is a framework to perform information extraction and integration of heterogeneous data sources

Semantic Integration of Information A common data model ODLI3 derived from

ODL-ODMG : standard languages for Object DataBasesI3 (Intelligent Information Integration) is a project of the ARPA (Advanced Research Project Agency) with the objective to define the reference architecture for automatic integration of heterogeneous sources

Tool-supported techniques to construct the Global Virtual View Local Schema Annotation w.r.t. a common lexical ontology

(WordNet) Semi-automatic generation of relationships between local

schemata Clustering techniques Semi-automatic generation of mappings between the GVV

and sources

MOMISMOMIS

Page 8: The MOMIS approach for  Information Integration

Domenico Beneventano8

DB

Gro

up @

uni

mo

MOMIS: Overview of the GVV-generation MOMIS: Overview of the GVV-generation process process

SYNSET2

SYNSET#

SYNSET4

SYNSET1

AUTOMATIC/MANUALANNOTATION

SEMI-AUTOMATICANNOTATION

INFERRED RELATIONSHIPS

LEXICON DERIVEDRELATIONSHIPS

SCHEMA DERIVEDRELATIONSHIPS

CommonThesaurus

COMMON THESAURUSGENERATION

USER SUPPLIEDRELATIONSHIPS

ODLI3LOCAL SCHEMA N

WRAPPING

ODLI3LOCAL SCHEMA 1

GVV GENERATION

MAPPING TABLES

GLOBAL CLASSES

clustersgeneration

Structuredsource

RDB

<XML>

<DATA>

Semi-StructuredSource

Page 9: The MOMIS approach for  Information Integration

Domenico Beneventano9

DB

Gro

up @

uni

mo

Person(Name, address)

Professor(Name, belongs_to,rank) FK: Name REF. Person FK: belongs_to REF. Division

Student (Name, year,e-mail) FK: Name REF. Person

Takes(Name, CourseName) FK: Name REF. Student FK: CourseName REF. Course

class Person(key Name){ attribute string Name; attribute string Name; }

class Professor : Person{ attribute Division belongs_to; attribute string rank; }

class Student: Person{ attribute set <Course> takes; attribute string rank; }

class Course{ … }

• Professor IS-A Person • Properties inheritance

Wrapping: from relational to objectWrapping: from relational to objectUNI (University Source) : relational source (Relational DB) ODLI3 schema

Page 10: The MOMIS approach for  Information Integration

Domenico Beneventano10

DB

Gro

up @

uni

mo

Interface Researcher{ attribute string name; attribute string e-mail; attribute set <Article> articles;}

Interface Teaching{ attribute string denomination; attribute string description;}

Interface Class : Teaching{ attribute string name; attribute integer year; attribute string specification; attribute Researcher professor;}…

Example of object sourceExample of object sourceCS (Computer Science Source) : object source (Object DB)

• complex object attributes: articles is a set of Article

• object attributes: professor is a researcher

Page 11: The MOMIS approach for  Information Integration

Domenico Beneventano11

DB

Gro

up @

uni

mo

Wrapping of Web: from pages to dataWrapping of Web: from pages to data

• A large number of Web sites contain now highly structured regions– Large web sites are generated automatically (statically or

dynamically) – Regularities occur at the page level, and at the site level as

well

• We aim at leveraging these regularities in order to face the issues of extracting fine grain data at a large scale

WrapperWrapper

xml documents or relational DB

Writing and maintaining wrappers is a labor intensive (and expensive) task

Road Runner Project (Università di Roma Tre & della Basilicata)

http://www.dia.uniroma3.it/db/roadRunner/

Page 12: The MOMIS approach for  Information Integration

Domenico Beneventano12

DB

Gro

up @

uni

mo

Wrapping of Web pagesWrapping of Web pages• Roadrunner: a tool to automatically generate wrappers for

web pages– Target: Data-intensive Web sites

• HTML pages generated by scripts that extract data from a large database

• Page class: collection of pages generated by the same script => similar structure

• Main Idea: exploit similarities and differences to infer a data extraction program (wrapper)

RoadrunnerWrapper generator

RoadrunnerWrapper generator

Input sample pages

Output wrapperOutput wrapperOutputdata

Pages similar in structure to those of the sample set

Page 13: The MOMIS approach for  Information Integration

Domenico Beneventano13

DB

Gro

up @

uni

mo

Set of terminological (intensional) relationships between terms (classes and attributes names) to express both intra-schema and inter-schema knowledge:

< Ti SYN Tj > Synonymy < Ti BT Tj > Broader Terms (BT) : hypernymy

< Ti NT Tj > (Narrower Terms - NT) hyponymy

< Ti RT Tj > (Related Terms - RT) positive association or holonomy

Common Thesaurus Generation:q lexicon derived relationships q schema derived relationshipsq designer supplied relationshipsq inferred relationships

Common ThesaurusCommon Thesaurus

Page 14: The MOMIS approach for  Information Integration

Domenico Beneventano14

DB

Gro

up @

uni

mo

Schema-derived RelationshipsSchema-derived Relationships

automatic extraction of intra schema relationships from each schema separatelyo BT from IS-A Relationships

< CS.TEACHING BT CS.CLASS>

o BT from relational foreign keys among primary keys

< UNI.PERSON BT UNI.PROFESSOR>

o RT from object attributes

< CS.CLASS RT CS.RESEARCH>

o RT from relational foreign keys

< UNI.PROFESSOR RT UNI.DIVISION>

Page 15: The MOMIS approach for  Information Integration

Domenico Beneventano15

DB

Gro

up @

uni

mo

NT

Hyponymy

Word Form Meaning (synset) teaching course … class education imparted in a series of lessons or class meetings

√ √

activities that impart knowledge

the profession of a teacher

Annotation and Lexicon-derived Annotation and Lexicon-derived RelationshipsRelationships

<UNI.COURSESYN CS.CLASS><CS.COURSE NT UNI.TEACHING>

Lexicon-derived relationships:Lexicon-derived relationships:

Lexicon relationships among terms of the sources can be derived on the basis of the semantic relationships between synset in WordNet (Hyponymy / Hypernymy, Meronymy, Correlation (between synsets having the same Hypernym), …) :

LOCAL SOURCE ANNOTATION : To assign meanings of class and attribute names w.r.t. a common lexical ontology (WordNet)

Page 16: The MOMIS approach for  Information Integration

Domenico Beneventano16

DB

Gro

up @

uni

mo

Extending WordNet: WordNet EditorExtending WordNet: WordNet Editor

If a class or attribute name has no correspondent in WordNet, the designer may add a new meaning and proper relationships to the existing meanings.

The designer may add a new meaning (for an existing word-form or for a new one) by: writing the gloss explicitly, or using an existing synset chosen among a list of

candidates obtained by an explicit search (using one or more keywords) or by exploiting similarity search techniques.

The designer may add relationships for the new synset Related synsets are obtained by an explicit

search (using one or more keywords) or by exploiting similarity search techniques.

Page 17: The MOMIS approach for  Information Integration

Domenico Beneventano17

DB

Gro

up @

uni

mo

Relationships derived as a logical consequence from a set of other Relationships. As a simple example, from

<Researcher RT Department> and < Researcher BT Professor>

we can derive that : <Professor RT Department>

Inferred RelationshipsInferred Relationships

The inferences are based on Description Logics techniques

Description Logics (DLs), describe knowledge in terms of concepts and their roles.

DLs provide a means of describing structured knowledge in a way that one can reason about it efficiently (i.e., infer new knowledge).

DLs are subsets of the First-Order Logic, providing decidability and computationally feasible reasoning at the cost of expressiveness.

ODB-Tools : DLs techniques for Object DataBase (www.dbgroup.unimo.it/ODB-Tools.html):

Page 18: The MOMIS approach for  Information Integration

Domenico Beneventano18

DB

Gro

up @

uni

mo

Clustersgeneration

Affinityevaluation

Common Thesaurus

Clusters of Classes

Affinity evaluation among classes: Name Affinity : computed by exploiting

relationships in the Common Thesaurus; a weight (W) is assigned to each kind of relationships, with WSYN >= WBT >= WRT

Structural Affinity: measures the affinity of class attributes

Global Virtual View GenerationGlobal Virtual View GenerationAffinity evaluation and Clusters Affinity evaluation and Clusters

GenerationGeneration

Clusters generation : Classes with affinity are grouped together in clusters by employing a

hierarchical clustering technique.

ARTEMIS tools (Università di Milano) https://islab.dico.unimi.it/artemis/.

Page 19: The MOMIS approach for  Information Integration

Domenico Beneventano19

DB

Gro

up @

uni

mo

Research_S

taff

Research_Staff

CS_Person

School_Member

Professor

Student

Univ_Student

...

CS

_Person

School_M

ember

Professor

Student

Univ_S

tudent

Section

Course ...

- 0.54 0.45 0.42 0.43

- - 0.6

0.68 0.68 0.56

- - - 0.45 0.65 0.6

- - - - 0.62 0.43

- - - - - 0.51

- - - - - -

0.36 0.39

0.2 0.2

0.2 0.2

0.25 0.25

0.25 0.25

0.2 0.2

0.5

Affinity MatrixAffinity Matrix

Page 20: The MOMIS approach for  Information Integration

Domenico Beneventano20

DB

Gro

up @

uni

mo

Affinity Matrix

Clustering Algorithm

Clustering Algorithm: (Hierarchical clustering techniques - Heveritt

74)

Cluster GenerationCluster Generation

Section Course

Room

CS_Person ProfessorStudent

School_Member

Univ_Student

Research_Staff

Division Department

Location

CL2

CL1

CL3

CL5

CL40.66

0.39

0.39

0.375

0.25

0.6

0.540.6

0.65

0.68

0.68

Page 21: The MOMIS approach for  Information Integration

Domenico Beneventano21

DB

Gro

up @

uni

mo

Common Thesaurus& Clusters

Global Virtual View

Mapping Table

generation

Global Classes generation :A global class C=(L,GA) is generated for each cluster : L are the local classes of the cluster GA are the global attributes of C

o Union of the local attributeso Fusion of “similar attributes” (by using the

Common Thesaurus)

Global Virtual View GenerationGlobal Virtual View GenerationGlobal Classes and Mapping Table Global Classes and Mapping Table

GenerationGeneration

Mapping generation : For each global class C=(L,GA), a Mapping Table (MT) is generated, to represent the mappings between global and local attributes MT is a table GAXL : An element MT[GA][L]

represents the attributes of the local class L mapped into the global attribute GA.

Global Classes and Attributes

generation

Page 22: The MOMIS approach for  Information Integration

Domenico Beneventano22

DB

Gro

up @

uni

mo

GVV and MT generation : exampleGVV and MT generation : example

CS.Teaching UNI. Course CS.Class Name denomination name name Description description specification Year year year Period period Professor professor

C= {UNI.Course, CS.Teaching, CS.Class}C= {UNI.Course, CS.Teaching, CS.Class} Cluster

Mapping Table of C

MT generation : Since CS.denomination NT UNI.Name, these local attributes correspond to the same global attribute GA

GVV annotation : the name and the meaning of the class C corresponds to the

name and the meaning of CS.Teaching (the broadest class) the name and the meaning of GA corresponds to the name and

the meaning of CS.description

Page 23: The MOMIS approach for  Information Integration

Domenico Beneventano23

DB

Gro

up @

uni

mo How to model the mappings between How to model the mappings between

the sources and the GVV?the sources and the GVV?

Is the GVV defined in terms of the sources? Approach global-schema-centric, or global-as-view (GAV)

Are the sources defined in terms of the GVV? Approach source-centric, or local-as-view (LAV)

A mixed approach? Approach called GLAV

GVV :

Teaching(Name,Description,Year,Period,Professor)

Sources :CS.Teaching(Denomination,Description)UNI.Course(Name,Year,Period)

CS.Class(Name,Specification,Year,Professor)

Page 24: The MOMIS approach for  Information Integration

Domenico Beneventano24

DB

Gro

up @

uni

mo

Local-as-ViewLocal-as-View

Local-as-View (LAV) mappings: associated to source classes we have views over

the GVVGVV :

Teaching(Name,Description,Year,Period,Professor)

Sources :CS.Teaching(Denomination,Description)UNI.Course(Name,Year,Period)CS.Class(Name,Specification,Year,Professor)

LAV mappings :CS.Teaching --> SELECT Name,Description FROM

TeachingUNI.Course --> SELECT Name,Year,Period FROM

TeachingCS.Class --> SELECT Name,Description

Year,ProfessorFROM Teaching

Page 25: The MOMIS approach for  Information Integration

Domenico Beneventano25

DB

Gro

up @

uni

mo

Global-as-ViewGlobal-as-View Global-as-View (LAV) mappings:

associated to classes in the GVV we have views over the sourcesGVV :

Teaching(Name,Description,Year,Period,Professor)

Sources :CS.Teaching(Denomination,Description)UNI.Course(Name,Year,Period)CS.Class(Name,Specification,Year,Professor)

GAV mappings :Teaching --> SELECT

f1(CS.Teaching.Denomination,…) AS Name, f2(…) As Description, …

FROM CS. Teaching FULL JOIN UNI.Course ON (…)

FULL JOIN CS.Class ON (…)

Page 26: The MOMIS approach for  Information Integration

Domenico Beneventano26

DB

Gro

up @

uni

mo GAV and LAV comparison (from Lenzerini GAV and LAV comparison (from Lenzerini IJCAI 2003)IJCAI 2003)

Page 27: The MOMIS approach for  Information Integration

Domenico Beneventano27

DB

Gro

up @

uni

mo

Global-as-View: the MOMIS approachGlobal-as-View: the MOMIS approach

• Global-as-View (GAV) mappings: for each global class C of the GVV we must define a conjunctive query QC over the local classes of C.

• GAV mappings: the MOMIS approachStarting from the Mapping Table of C, the integration designer, supported by the Ontology Builder graphical interface, can implicitly define QC by:

• using and extending the Mapping Table with• Data Conversion Functions from local to global

attributes• Join Conditions among pairs of local classes belonging

to C• Resolution Functions for global attributes to solve

data conflicts of local attribute values.

– using and extending the Full Disjunction operator

Page 28: The MOMIS approach for  Information Integration

Domenico Beneventano28

DB

Gro

up @

uni

mo

Building the Mappings: an exampleBuilding the Mappings: an example

StringConc (REGION,STATE)

from T_L1 full join T_L2

Data Conversion Functions

Join Attribute

on (T_L1.COMPANY_ID = T_L2.COMPANY_ID)

Join Conditions

FullDisjunction

L1.company L2.company

COMPANY_ID COMPANY_ID COMPANY_ID

SUBCONTR SUBCONTR

CAPITAL_STOCK CAPITAL_STOCK

REGION REGION REGION, STATE

ADDRESS ADDRESS ADDRESS

...Select COMPANY_ID,

precedence(T_L1.ADDRESS, T_L2.ADRESS) as Address, T_L2.SUBCONTRACTOR, …

Resolution Functions

Precedence(L1,L2)

Page 29: The MOMIS approach for  Information Integration

Domenico Beneventano29

DB

Gro

up @

uni

mo

Join Conditions and Full DisjunctionJoin Conditions and Full Disjunction• Object Identification: Merging data from different sources requires

different representations of the same real world object to be identified.

• Join Conditions: To identify instances of the same object and fuse them among pairs of local classes belonging to the same global class. As an example, for the global class Company the designer can define JC(L1.Company,L2.Company) :

L1.COMPANY_ID = L2.COMPANY_ID

• Full Disjunction (FD) “computing the natural outer-join of many relations preserving all possible connections among facts”

• Given a global class C composed of L1,L2, ..., Ln we consider

FD(T(L1), T(L2), . . . , T(Ln)) computed on the basis of the Join Conditions

where T(L) denotes L transformed by the Data Conversion Function, i.e., the full disjunction operator is applied after data conversion.

Page 30: The MOMIS approach for  Information Integration

Domenico Beneventano30

DB

Gro

up @

uni

mo

FD(L1,L2) : select S(L1) union S(L2) from L1 outer join L2 on JC(L1,L2)

Full Disjunction: an exampleFull Disjunction: an example

COMPANY_ID ADDRESS CAP_STOCK

123 Via Ugo 55

4 Via Po, 2 65

COMPANY_ID ADDRESS SUBCONTR

4 Via Larga 3 Yes

234 Via Verdi 9 No

L1.COMPANY_ID L1.ADDRESS L1.CAP_STOCK L2.COMPANY_ID L2.ADDRESS L2.SUBCONTR

123 Via Ugo 55

4 Via Po, 2 65 4 Via Larga 3 Yes

234 Via Verdi 9 No

L1=L1.Company L2=L2.Company

where JC(L1,L2) = L1.COMPANY_ID = L2.COMPANY_ID

Page 31: The MOMIS approach for  Information Integration

Domenico Beneventano31

DB

Gro

up @

uni

mo

Resolution FunctionsResolution Functions

• The fusion of data coming from different sources has to take into account the problem of inconsistent information among sources.

• In the context of MOMIS we adopt the Resolution Function. – A Resolution Function may be defined for each global

attribute mapping onto local attributes coming from several local sources, in order to solve data conflicts due to different local attribute values.

– Homogeneous Attributes : If there are no data conflicts for a global attribute mapped onto more than one source

– As an example, in GVV.Company, we define all the global attributes as Homogeneous Attributes except for Address where we used a precedence function:

L1.ADDRESS has a higher precedence than L2.ADDRESS

Page 32: The MOMIS approach for  Information Integration

Domenico Beneventano32

DB

Gro

up @

uni

mo

Resolution FunctionsResolution Functions

COMPANY_ID ADDRESS CAP_STOCK SUBCONTR

123 Via Ugo 55

4 Via Po 2 65 Yes

234 Via Verdi No

L1.COMPANY_ID L1.ADDRESS L1.CAP_STOCK L2.COMPANY_ID L2.ADDRESS L2.SUBCONTR

123 Via Ugo 55

4 Via Po 2 65 4 Via Larga 3 Yes

234 Via Verdi No

Application of the resolution functions

Page 33: The MOMIS approach for  Information Integration

Domenico Beneventano33

DB

Gro

up @

uni

mo

FD Computation with more that 2 classesFD Computation with more that 2 classes• [Rajarama, Ullman - PODS 1996] : There is a outerjoin sequence

producing FD if and only if the set of relation forms a connected, acyclic hypergraph.

A Global Class C with more than 2 local classes is a cyclic hypergraph we need a new method

Example with n = 3 :

L1

L2 L3

JC(L1,L3)JC(L1,L2)

JC(L2,L3)

• The computation of FD is performed assuming: 1. each L contains a key, 2. all the join conditions are on key attributes, 3. all the join attributes are mapped into the same set of global attributes (K).

Then, FD can be computed as: L1 full join L2 on JC(L1,L2))

full join L3 on (JC(L1,L3) OR JC(L2,L3)

Page 34: The MOMIS approach for  Information Integration

Domenico Beneventano34

DB

Gro

up @

uni

mo

Global Query ManagementGlobal Query Management The querying problem:

How to answer queries expressed on the GVV (global queries)?

Query rewriting : to rewrite a global query as an equivalent set of queries expressed on the local schemata (local queries).

GAV approach: the query is processed by means of unfolding :

by expanding a query on a global class C of the GVV according to the definition of QC

Fusion and Reconcilation of the local answers into the global answer

Object Fusion & Resolution Functions

Inconsistencies between sources

Query Optimization

Page 35: The MOMIS approach for  Information Integration

Domenico Beneventano35

DB

Gro

up @

uni

mo

Q: SELECT COMPANY_ID, REGION, ADDRESS FROM company WHERE CAPITAL_STOCK > 50 AND REGION LIKE '*VENETO*' AND SUBCONTR LIKE ’yes’

LQ1SELECT COMPANY_ID, REGION,ADDRESS, FROM L1.company WHERE (REGION like '*VENETO*' and CAPITAL_STOCK > 50)Local

queries

Query unfolding exampleQuery unfolding example

Global Query

Global Class: Company = { L1.Company, L2.Company}

LQ2SELECT COMPANY_ID, REGION,STATE,ADDRESS, FROM L2.company WHERE ( strconc(REGION,STATE) like '*VENETO*' and

SUBCONTR like 'yes')

Page 36: The MOMIS approach for  Information Integration

Domenico Beneventano36

DB

Gro

up @

uni

mo

Query unfolding exampleQuery unfolding example

Full disjunction computation: FD : Select *

from T_LQ1 full outer join T_LQ2on (T_LQ1.COMPANY_ID = T_LQ2.COMPANY_ID)

Query Answer Select COMPANY_ID, REGION,

precedence(T_LQ1.ADDRESS, T_LQ2.ADDRESS) AS ADDRESSfrom RP

Residual predicate: RP: SELECT *

FROM FD WHERE CAPITAL_STOCK > 50 AND SUBCONTR LIKE ’yes’

Q: SELECT REGION, ADDRESS FROM company WHERE CAPITAL_STOCK > 50 AND REGION LIKE '*VENETO*' AND SUBCONTR LIKE ’yes’