1 ©keith g jeffery, anne assersoncris and dataspacescris2010, aalborg cris and dataspaces by...

1©Keith G Jeffery, Anne Asserson CRIS and Dataspaces CRIS2010, Aalborg

CRIS and DataSpaces by Epitaxial Growth

Keith G Jeffery STFCAnne Asserson UiB With acknowledgement to

http://www.mineral-forum.com/


Authors

Anne Asserson UiB

Keith G Jeffery STFC-RAL


Structure• History

• Hypothesis

• Achievement

• Conclusion

– CRIS Interoperation Requirement

– Classical Interoperation– CRIS Interoperation– Dataspaces

– Requirement– Challenges– Solution Concept

– Schema matching and mapping

– Other Challenges


History: CRIS Interoperation Requirement

• Internationally and Nationally– (a) to allow each funding organization to make strategic funding

decisions related to knowledge of the actions of other funding organisations; (b) to provide an international base of reviewers for research proposals;

– (c) to provide comparative metrics on research performance and cost-benefit;

– (d) for researchers to find colleagues in areas peripheral to their own (where they claim they knew the key players) on an international basis;

– (e) for encouraging international innovation taking research outputs through to wealth-creation.

• For business purposes– Funding organisations <-----> research organisations


History: Classical Interoperation Technology

• Significant challenge in ICT / Computer Science– For structured data sources

• From formal schemas erect global schema to reduce n(n-1) interoperations to n

– For semi-structured data sources• From XML schema as above • else derive schema from tags

– For unstructured data sources• Harvesting • Possibly advanced knowledge engineering and machine

learning techniques• All require considerable human intervention• This does not scale for future internet


History: CRIS Interoperation Technology

• IDEAS (1984-87), EXIRPTS (1987-89), ERGO 1997-99)

• Categorisation of Techniques (2005)– Remote Wrapper– Local Wrapper– Catalog– Catalog Plus Pull (ERGO2++)– Full CERIF– Harvesting

• Confirmed by euroHORCs special group 2008– Recommended converge to CERIF to allow evolution

to Full CERIF technique.


History: Dataspaces• The vision behind dataspaces is that the end-user ‘sees’ a space

with all relevant information required for a particular purpose presented in a form suitable for the purpose of the end-user.

• Utilise human effort and knowledge to guide (or actually execute) the creation of a temporary, partial global schema by matching of schemas and mapping (i.e. specifying the wrapper software) for interoperation.

• More recently it has been suggested that machine learning could be employed using manually-created mappings as training data.

• Dataspaces includes the concept of ‘pay as you go’ only requiring the matching and mapping necessary for a particular interoperation instance.

• The incremental nature of building / accessing dataspaces stimulated the concept of epitaxial growth in dataspaces



• Hypothesis

• Achievement

• Conclusion







Hypothesis: Requirement

Local CRIS

remote CRIS

remote CRIS

remote CRIS

query results

CERIF

CERIF

CERIF


Hypothesis: Challenges1. discovery of relevant CRIS;2. description of each relevant CRIS;3. matching each description in (2) to CERIF including

(a) management of different character sets and multimedia representation;(b) management of different languages;(c) management of different syntax (data structure); (d) management of different semantics (meaning);

4. generating mappings to describe the required conversions on instances under each CRIS schema to/from CERIF;

5. generating conversion software for instances in any source CRIS;6. managing a query in terms of a local schema (or CERIF) over all

CRIS;7. managing the response (answers);8. management of change in schemas (3) (but usually only (c) and (d));


Hypothesis: Solution Concept

With acknowledgement to



Epitaxial Growth

Structurally congruent

The term epitaxy comes from the Greek roots epi, meaning "above", and taxis, meaning "in ordered manner".

ions

Crystal

structure


Hypothesis: Solution Concept

project

person organisation

‘seed’schema

extended project schema

extendedpersonschema

extendedorganisation

schema

source 1 source 2

source 1source 2 source 1

source 2

input

epitaxial

growth

The Concept of Epitaxial Growth of Schemas



• Hypothesis

• Achievement

• Conclusion







Achievement: Schema Matching and Mapping

• Challenges 3 and 4 (matching and mapping) are addressed using the novel technique

• We propose epitaxial growth from a canonical CERIF schema to generate the global schema for any group of CRIS.

• In the process, for each CRIS, there is documented – entities and attributes matching CERIF (CRIS=CERIF); – entities and attributes in CERIF that are missing in the CRIS

CERIF > CRIS; – entities and attributes in the CRIS that are missing in CERIF

CRIS > CERIF;


Achievement: Schema Matching and Mapping

– lAnguage match;– Lexical match;– sYntactic match; – sEmantic match;

– Reconciliation;– Epitaxial growth;

Schemas of CRIS and CERIF represented as fully connected cyclic graphs.

node

A

LY

E

Match Flags


Achievement: Match FlagsPhase E type T type M =

MatchR = paRtial match

N = No match

A = lAnguage AM AR AN

L = Lexical LM LR LN

Y = sYntactic YM YR YN

E = sEmantic D = Dictionary EMD ERD END

T = Thesaurus B = suBterm EMTB ERTB ENTB

P = suPerterm EMTP ERTP ENTP

O = Ontology EMO ERO ENO


Achievement: IdealPhase E type T type M =


N = No match









Achievement: Translation

• If Flags AR | AN • To reach AM• Utilise E : Semantic

analysis• To achieve

– EMD– EMTB– EMTP– EMO

node

A

LY

E


Achievement: TranslationPhase E type T type M =


N = No match









Achievement: Improvement• If Flags

– LR | LN• To reach LM• Utilise

– Y: Syntactic analysis for structural positioning to achieve

• YM– E: Semantic analysis for

related terms to achieve • EMD• EMTB• EMTP• EMO

node

A

LY

E


Achievement: ImprovementPhase E type T type M =


N = No match









Achievement: Algorithm (sketch) 1• AM | LM | YM | [EMD|EMO] direct mapping

– the issue of EMTB, EMTP is dealt with below;• In all other cases R | N may be resolved by a

later phase;• AR | AN flag precludes lexical match unless

resolved by ED | ET | EO;• AM + LR | LN flag precludes a match unless

resolved by ED | ET | EO; • Y processing indicates nodes to attempt to

match using ED | ET | EO;


Achievement: Algorithm (sketch) 2

• ED provides for multilingual term equality;• ET processing like Y; the processing must cycle

over the graph to find the closest YM and then apply ET to find EMTP | EMTB and allocate the flag(s) appropriately;

• EO processing like Y; the processing must cycle over the graph to find the closest match for the current node by locating EMO for terms in nodes YM-related to the current node thus determining the semantics of the current node;


Achievement: Epitaxial Growth




Achievement: Epitaxial Growth

• In the general case we now have two graphs with matching and non-matching nodes.

• likely there are excess nodes in the CRIS graph (i.e. the CRIS contains all the CERIF elements, probably with different names, plus more entities / attributes suitable for local processing)


Achievement: Epitaxial GrowthCRIS Superset

e

e a a

a a e a a

a aa

e

a e

a

a

e

a

a

e

e

e a a

a a e a a

a aa

e

a e

a

a

e

a

a

e

a a

CRIS CERIF

e = entity, a = attribute


Achievement: Epitaxial GrowthCRIS Subset

e

e a a

a a

a

e

a e

a

a

e

a

a

e

e a a

a a

a

e

a

a a

CRIS CERIF

e = entity, a = attribute


Achievement: Epitaxial Growth: Algorithm (sketch) 1

• For nodes in the CRIS schema that do not have corresponding nodes in the CERIF schema:

• Their syntactic positions in the graph, related to nodes in the CRIS schema, are analysed

– starting from the root and working down the graph, left to right at each level;

• The language is flagged– so that this can be the variable value input in the CERIF language

attribute;• The lexical term for the entity or attribute name is noted ready for

translation to the canonical language for interoperation – with the CERIF translation attribute set to automatic;

• The syntactic structure present in the CRIS but not in CERIF is then added to the CERIF schema as the epitaxial growth……...



• This is achieved in the following steps:– if the additional nodes are attributes under an

existing entity ‘X’, a new CERIF entity ‘Xadditional’ is created with the attributes as nodes under it and ‘Xadditional’ is linked with a cardinality 1:1 with ‘X’ in CERIF using matched primary keys;


Achievement: Epitaxial GrowthAdditional Attributes

CERIF Entity

‘X’

NEWCERIF Entity

‘Xadditional’

NEWCERIF LinkingRelation

‘X’‘Xadditional’

CRIS‘Y’

CRISEntity‘Y=X’

CRISEntity‘Y≠X’



– if the additional nodes include an additional entity ‘Y’, linking relations are created in CERIF to link ‘Y’ to existing CERIF entities and attributes of ‘Y’ are nodes under ‘Y’;

– This step depends on the complexity of the CRIS: primary/foreign key relationships (with cardinality 1:n) can be analysed to generate the linking relations and if the CRIS already has n:m cardinality relationships using linking relations these can be mapped directly;


NEWCERIF Linking

Relations‘Y’

To other entities

NEWCERIF Linking

Relations‘Y’

To other entities

Achievement: Epitaxial GrowthAdditional Entity

NEWCERIF Entity

‘Y’

NEWCERIF Linking

Relations‘Y’

To other entities

CRIS‘Y’

CRISEntity‘Y≠X’

For all X

Links to other entities



• A similar process is followed for nodes in the CERIF schema not having corresponding nodes in the CRIS schema.


Achievement: Other Challenges• Challenge 1: DISCOVERY

– managed by the CRIS community knowledge (e.g. DRIS) and/or by web services searches UDDI/WSDL.

• Challenge 2: DESCRIPTION– the schema of any structured (or semi-structured) CRIS.

• {Challenges 3 and 4 : MATCHING and MAPPING– described above}

• Challenge 5: CONVERSION– using technology from Hypermedata (SkKoBeJe1999). From the

schema graphs of the CRIS and CERIF the convertor can be generated to convert instances under one schema to instances under another.

• Challenges 6 and 7; QUERY and RESPONSE– using technology from MIPS (JeHuKaWiBeMa94)

• Challenge 8; CHANGE– managed by repeating the above as required



• Hypothesis

• Achievement

• Conclusion







Conclusion 1• questions facing society today can be solved, or partially solved

using research information to guide actions and developments. • world-scale problems cannot be solved by local-scale research

information.• base technology to support and move forward is interoperating

(CERIF-)CRIS within a (European) e-infrastructure.• current technologies do not scale to ‘future internet’.

• responsibility of the CRIS community to find, develop and promote a solution for interoperating CRIS that is reliable, scalable, economic and has appropriate security and privacy.

• interoperating CRIS using epitaxial growth is a promising line of development.


Conclusion 2

And of course, it would be much more effective and efficient……..

……..If all research information systems used CERIF as their storage and interoperation format

1 ©keith g jeffery, anne assersoncris and dataspacescris2010, aalborg cris and dataspaces by...

Documents