Linking Big Data to Rich Process Descriptions

Introduction Motivation Math. Semantics Ontology Integration & Interoperability Software Verification Conclusion

Linking Big Data toRich Process Descriptions

Christoph Lange1

1Project ‘‘Formal Mathematical Reasoning in Economics’’,School of Computer Science, University of Birmingham, UK

http://cs.bham.ac.uk/~langec

2013-09-19

Lange Linking Big Data to Rich Process Descriptions 2013-09-19 1

http://cs.bham.ac.uk/research/projects/formare/

http://cs.bham.ac.uk/~langec


“Hello, World!”

2011 Ph.D. (Jacobs Univ. Bremen, with MichaelKohlhase): Enabling Collaboration onSemiformal Mathematical Knowledge bySemantic Web Integration [Lan11]

2011–12 Postdoc (Univ. Bremen, with John Bateman,Till Mossakowski): Ontology Integration andInteroperability (OntoIOp)↝ DistributedOntology Language (DOL) OMG standard [13]

2012–13 Postdoc (Univ. Birmingham, with ManfredKerber, Colin Rowat): Formal MathematicalReasoning in Economics (ForMaRE) [KLR]



Linking Big Data toRich Process Descriptions

Linked data is one key to coping with big data.Big data are created or consumedby technical/business processes.Formal process descriptions aremore complex than data.Why integrate process descriptions and data?How to integrate them?



Linked Data as a Key to Big Data

Linked (Open) Data enables . . .decentralised, collaborativemanagement of bigdatasets,low-overhead information retrieval, andscalable reasoning.



Sources of Big Data

Big data =

)︀⌉︀⌉︀⌉︀⌋︀⌉︀⌉︀⌉︀]︀

high volume,high velocity,high variety

[︀⌉︀⌉︀⌉︀⌈︀⌉︀⌉︀⌉︀⌊︀information [BL12]

Where does it come from?

Science 150 million sensors in the Large HadronCollider

Trade High-frequency trading (HFT) accounts for50% of US equity trading.

Web 100 hours of video uploaded to YouTubeevery minute



Big Data Result from Processes

Science sensor measurements determined byexperimental setupexperiments inform hypotheses

Trade trading strategies influenced bydemand and supply

Web YouTube does not just store uploads,but notifies

subscribers,Facebook friends,Twitter followers.



Formal Process DescriptionsWhy describe processes formally?

to check their compliancewith quality standardsto verify the software that controls them

Science workflows modelled using:logic programming, computational tree logic,linear temporal logic.

Trade Knight Capital HFT software repeatedly soldshares below purchase price, lost $440million within 1 hour – could formalverification have helped?

Web social networks modelled usingepistemic modal logic, probabilistic soft logic



Linking Process Descriptions and Data

Knowledge Mgmt. Under what experimental setupwere these measurements taken?

Reasoning Given the current variance ofmeasurements, would it help to use asensor with different specifications?Which trading strategy responds bestto the current offers?

Inform. Retrieval Which of my friends are actuallyinterested in my latest video upload?Where can I buy the cheapest parts tofeed into my manufacturing process?



Logic of Linked Open Data?

RDF data and RDFS vocabularies do not suffice formodelling processes – so . . . ?

☀ make your stuff available on the Web(whatever format) under an open license☀☀ make it available as structured data (e.g.,Excel instead of image scan of a table)☀☀☀ use non-proprietary formats (e.g., CSVinstead of Excel)☀☀☀☀ useURIs to denote things, so that peoplecan point at your stuff☀☀☀☀☀ link your data to other data to providecontext [12]

Who says it needs to be RDF?



Logic of Linked Open Data?

RDF data and RDFS vocabularies do not suffice formodelling processes – so . . . ?

☀ make your stuff available on the Web(whatever format) under an open license☀☀ make it available as structured data (e.g.,Excel instead of image scan of a table)☀☀☀ use non-proprietary formats (e.g., CSVinstead of Excel)☀☀☀☀ useURIs to denote things, so that peoplecan point at your stuff☀☀☀☀☀ link your data to other data to providecontext [12]

Who says it needs to be RDF?Lange Linking Big Data to Rich Process Descriptions 2013-09-19 9


Think URIs, not RDF!

How to achieve an integration of . . . ?rich process descriptions (expressive logics)big data (scalability before expressivity)

Ad hoc extensions of RDF exist (e.g. for CSPs in productrange specification at Renault [BSP11])

My approachsystematically base expressive logics beyond RDFand OWL on the URI foundation of LODthus enable large-scale data/knowledge integration



Derived Values in Statistical Datasets

Comparing unemployment rates in micronations:

Principality of Sealand

:pop_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 7 .

:unemployed_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 2 .

:unemp_rate_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;sdmx-attr:unitMeasure :Ratio ;sdmx-meas:obsValue 0.286 .

Republic of Kugelmugel

:pop_kugelmugel2012 a qb:Observation ;sdmx-dim:refArea :KugelmugelRepublic ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 11 .

:unemployed_kugelmugel2012 a qb:Observation ;sdmx-dim:refArea :KugelmugelRepublic ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 1 .

:unemp_rate_kugelmugel2012 a qb:Observation ;sdmx-dim:refArea :KugelmugelRepublic ;sdmx-dim:refPeriod :Year2012 ;sdmx-attr:unitMeasure :Ratio ;sdmx-meas:obsValue 0.091 .



Derived Values in Statistical Datasets

Comparing unemployment rates in micronations:

Principality of Sealand

:pop_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 7 .

:unemployed_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 2 .

:unemp_rate_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;sdmx-attr:unitMeasure :Ratio ;sdmx-meas:obsValue 0.286 .

Republic of Kugelmugel

:pop_kugelmugel2012 a qb:Observation ;sdmx-dim:refArea :KugelmugelRepublic ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 11 .

:unemployed_kugelmugel2012 a qb:Observation ;sdmx-dim:refArea :KugelmugelRepublic ;sdmx-dim:refPeriod :Year2012 ;:refAgeGroup :People18to65years ;sdmx-attr:unitMeasure :Count ;sdmx-meas:obsValue 1 .




Derived Values in Statistical Datasets II:unemp_rate_sealand2012 a qb:Observation ;sdmx-dim:refArea :PrincipalityOfSealand ;sdmx-dim:refPeriod :Year2012 ;sdmx-attr:unitMeasure :Ratio ;sdmx-meas:obsValue 0.286 .


How to validate the derived values?How to compute them for new data points?How to collect data points and their dependencies?

Make themathematical semantics explicit!unemp. rate = unemployed

population ⇒ link to “division”(→OpenMath Content Dictionaries) [Vra+10; Lan10]OpenMath CDs are LOD: decentrally extensibleFuture work: OpenMath SPARQL entailment regime



The Big Picture of Interoperability

Ontology

Ontology Language/Logic

Knowledge Software Agents

written in

Concepts/Data/Individuals

represented in terms of

Service Description

Service Descr. Language

written in

Service

satisfies

processes

refers to

Target (Device)accesses

Service-Oriented Architecture

Smart Environment

Target Description

conforms to

Device

Target Descr. Language

written in

Ontology

Ontology Language/Logic

Concepts/Data/Individuals

Service Description

Service Descr. Language

Service Target (Device)

Target Description

Device

Target Descr. Language

Knowledge Infrastructure

map

ping

s fo

rin

tero

pera

bilit

y

Hardware

Data

Models

Metamodels

For now we focuson the “content”/“knowledge”column



Towards Device InteroperabilityAmbient Assisted Living ScenarioClara, a vegetarian, instructs her wheelchair to get her tothe kitchen (next door to the living room). For dinner,she would like to take a pizza from the freezer and bakeit in the oven. Afterwards she goes to bed.

Existing ontologies (e.g. OpenAAL) cover core of that:

. . . but not all required concepts (e.g. foodingredients⇒ need other ontologies/modules; tapinto the Web of (Product, Geo) Data). . . not necessarily at the required level ofcomplexity (e.g. space/time⇒ need other logics)


http://openaal.org


Towards Device InteroperabilityAmbient Assisted Living ScenarioClara, a vegetarian, instructs herwheelchair to get herto the kitchen (next door to the living room). Fordinner, she would like to take a pizza from the freezerand bake it in the oven. Afterwards she goes to bed.

Existing ontologies (e.g. OpenAAL) cover core of that:

. . . but not all required concepts (e.g. foodingredients⇒ need other ontologies/modules; tapinto the Web of (Product, Geo) Data). . . not necessarily at the required level ofcomplexity (e.g. space/time⇒ need other logics)


http://openaal.org



Existing ontologies (e.g. OpenAAL) cover core of that:. . . but not all required concepts (e.g. foodingredients⇒ need other ontologies/modules; tapinto the Web of (Product, Geo) Data)

. . . not necessarily at the required level ofcomplexity (e.g. space/time⇒ need other logics)


http://openaal.org



Existing ontologies (e.g. OpenAAL) cover core of that:. . . but not all required concepts (e.g. foodingredients⇒ need other ontologies/modules; tapinto the Web of (Product, Geo) Data). . . not necessarily at the required level ofcomplexity (e.g. space/time⇒ need other logics)


http://openaal.org


What do Devices Need to Know?

Some of the devices involved:kitchen light switchfreezer (aware of its contents)wheelchair (with navigation)

Services and Devices need to understand differentaspects of the world at different levels of complexity.

Quote from the “Hitchhiker”“Suddenly [the door] slid open.‘Thank you,’ it said, ‘for making asimple door very happy.’”



Different Devices and their Knowledge

Light Switch “switched on if and only if someone is inand it’s dark outside”

Freezer “all toppings of a vegetarian pizza arevegetarian”

Wheelchair “two areas are either the same, or intersect,or border, or separate, or one is part of theother”



Different Devices = Different LogicsLight Switch: propositional logic “switched on if andonly if someone is in and it’s dark outside”light_on ≡ person_in_room ∧ dark_outsideFreezer: description logic (Pizza ontology) “alltoppings of a vegetarian pizza are vegetarian”VegetarianPizza ≡ Pizza ⊓ ∀hasTopping.VegetarianWheelchair: first order logic (RCC-style spatialcalculus) “two areas are either the same, or intersect, orborder, or separate, or one is part of the other”∀a1, a2.equal(a1, a2) ∨ overlapping(a1, a2) ∨bordering(a1, a2) ∨ disconnected(a1, a2) ∨part_of(a1, a2) ∨ part_of(a2, a1)



The OntoIOp Initiative

OntoIOp (Ontology Integration andInteroperability) initiative started in 2011 with ISO

now continued with OMGRequest for Proposals to be issued this autumnproposals due Dec. 2014

50 experts participate, ∼ 15 have contributedRelevant communities represented:

different ontology languages and logicsconceptual and theoretical foundationstechnical foundationsapplications: manufacturing, business rules,model-driven software engineering



Distributed Ontology Language (DOL)“distributed” means . . .

logically heterogeneousmodularinterlinked: interpretations, equivalences, alignmentsdecentrally maintained (using URIs)

DOL: a logic-agnostic meta-language forontologies, modeling and specification [MKL12;Lan+12]

supports ontologies in several relevant languagesframework can be decentrally extended with newlanguages, logics, serializations, translations

Tool support:Hets: syntax check, theorem proving, model findingOntohub: web-based repository engine

http://ontoiop.orgLange Linking Big Data to Rich Process Descriptions 2013-09-19 19

http://ontoiop.org


The OntoIOp Registry (Subset)

Common Logic

SROIQDL-LiteR

CLIF

XCL

Manchester Syntax

OWL 2 XML

RDF / XML

Turtle

OWL 2 DL

RDF

RDFS

Common Logic

RDFS

RDF

OWL 2 QL

OWL 2 RL

OWL 2 EL

DL-RL

EL++

Serializations Ontology Languages Logics

supports serialization sublanguage of

induced translation exact logical expressivity

translatable to

sublogic of



DOL AAL Example

%prefix( : <http://dfki.de/example/ouraal/> openaal: <http://openaal.org/SAM/Ontology#>productdb: <http://productdb.org/ean/> pizza: <http://www.co-ode.org/ontologies/pizza/pizza.owl#>lang: <http://purl.net/dol/languages/> log: <http://purl.net/dol/logics/>trans: <http://purl.net/dol/translations/> ser: <http://purl.net/dol/serializations/> )%

distributed-ontology AAL

language lang:OWL2/DLontology OpenAALAdapted =openaal: %(import_of_openaal)% with openaal:AP ↦ AssistedPerson

then %(some_extensions)% syntax ser:OWL2/Manchester {Class: LightSwitch SubClassOf: openaal:DeviceClass: Freezer SubClassOf: openaal:Device %(freezer_sub_device)%

Class: RoomWith1PersonEquivalentTo: openaal:Room that inverse openaal:is-in-room min 1 AssistedPerson

Class: RoomWithAllLightsOnEquivalentTo: openaal:Room that inverse openaal:is-in-room only

(not (LightSwitch that openaal:has-power-state value openaal:Off)) }

then logic log:Propositional syntax ser:Prop/CASLLike : {props light_on, person_in_room, dark_outside. light_on ⇔ person_in_room ∧ dark_outside }

with translationtrans:PropositionalToSROIQperson_in_room ↦ RoomWith1Person,light_on ↦ RoomWithAllLightsOn



DOL AAL Example












DOL AAL Example












DOL AAL Example












DOL AAL Example












DOL AAL Example












DOL AAL Example II

then language lang:OWL2/DL : pizza:

then logic log:CommonLogic syntax ser:CommonLogic/CLIF : {(forall (area1 area2)(or (equal area1 area2) (overlapping area1 area2) %% skipping some cases ...

%% (define mutual disjointness of these predicates)

(forall (area1 area2)(if (or (equal area1 area2) %% ...

(exists (door)(and (openaal:Door door)

(openaal:is-in-room door area1)(openaal:is-in-room door area2))))

(openaal:is-connected-to-room area1 area2))) }

ontology ConcreteScenario =OpenAALAdapted hide along trans:RDFtoSROIQand productdb:

then language lang:RDF syntax ser:RDF/Turtle : {productdb:4001724819806 pizza:hasTopping[ a pizza:TomatoTopping ], [ a pizza:MozzarellaTopping ] .

} with translation trans:RDFtoOWL2DLthen { pizza:then syntax ser:OWL2/Manchester : {Individual: productdb:4001724819806Types: pizza:hasTopping exactly 2 }



DOL AAL Example II













DOL AAL Example II













DOL AAL Example II













DOL AAL Example II













DOL AAL Example II













Importance of AuctionsAuctions: a mechanism to distribute resourcesApplications eBay, mobile spectrum, internet domainsSignificance $268.5 billion in 2008 in the US

Given a set of bids on goods (proxying valuations)Goals give goods to those valuing themmost

determine pricesmaximise revenueattract participantsincentive compatibility(no need for tactic over-/underbidding)

Auctions are designed; properties are tested andproved.



Generating Verified Auction Software

2. Theorems

1. Definitions

formal specification

(written by Isabelle user, needs

review by auction designer)

Code(executable Scala)

3. Proof(4. checked by Isabelle)

state soundness

and other properties of

known to implement (by proof

and by trusting code generator)

5. code generation

(Isabelle)

proves



Combinatorial Auctions [CSS06]



Combinatorial Vickrey AuctionBids on any subset of the set of available goods X .Winning allocation:

X∗ ∈ argmaxX1 ,...,XN

N

∑n=1bn (Xn) s.t.

N

⋃n=1Xn ⊆ X0, n ≠ n′ iff Xn∩Xn′ = ∅

Prices: pn ≡ αn −∑m≠n bm (X∗m)whereαn ≡ max

Xmm=1,...,N,m≠n

{∑m≠nbm (Xm) ⋁︀⋃m≠nXm ⊆ X0(︀

Bidder n pays the maximum sum of bids if the auctionhad been run without n (= αn), minus the winning bidson the items n did not get [AM06; Cam+13].



Generating Verified Software: Comb.Vickrey Auction [Cam+13]

paper-likeformalisation

X ∗ ∈ argmax∑ . . .

{R ⊆P(N)×N ∣∃P ∈ parts(G).Dom(R) ⊆ P∧ . . .}

{P ∣ ⋃P = A∧∀x ∈ P. . . .}

depends on

depends on

executableformalisation

argmax (x # xs) f =

if f x > f (hd (argmax xs f)) then ...

alloc G N = concat [

[ R . R ← inj_fun P (list N) ]

. P ← parts (list G) ]

parts (x # xs) =⋃ inject x ‘ (parts xs)

depends on

depends on

!≡winner

determination

!≡allocations

!≡set partitions

papers

ource(

auctiondesig

ner)

veri�e

dcode

(auctio

nso�w

are)

humanformali-sation

codegene-ration

http://formare.github.io/auctions/Lange Linking Big Data to Rich Process Descriptions 2013-09-19 27

http://formare.github.io/auctions/


Conclusion

Formal descriptions help to understand, verify andimprove processes in general.Process executions create or consume data.Integrating process descriptions and data improves

knowledge managementreasoninginformation retrieval

A wider view on linked data (beyond RDF) helps tointegrate . . .

process descriptions(often ≥ first-order logic; expressive)big data created or consumed by processes(often RDF; scalable)


References

References I

5 star Open Data. Apr. 3, 2012. url:http://5stardata.info/ (visited on 2013-09-18).

OntoIOp (Ontology Integration and Interoperability)Standard Development Initiative. 2013. url:http://ontoiop.org (visited on 2013-09-18).

L. M. Ausubel and P. Milgrom. “The Lovely but LonelyVickrey Auction”. In: Combinatorial auctions. Ed. byP. Cramton, Y. Shoham, and R. Steinberg. MIT Press,2006. Chap. 1, pp. 17–40.

M. A. Beyer and D. Laney. The Importance of ‘Big Data’:A Definition. June 21, 2012. url:http://www.gartner.com/resId=2057415.


http://5stardata.info/

http://ontoiop.org

http://www.gartner.com/resId=2057415

References

References IIF. Badra, F.-P. Servant, and A. Passant. “A SemanticWeb Representation of a Product RangeSpecification based on Constraint SatisfactionProblem in the Automotive Industry”. In: Proceedingsof the 1st Workshop on Ontology and Semantic Web forManufacturing, Extended Semantic Web Conference.(Hersonissos, Crete, Greece, May 29, 2011). Ed. byA. García Castro, C. Toro, L. Ramos, and L. Schröder.CEUR Workshop Proceedings 748. Aachen, 2011,pp. 37–50. url: http://ceur-ws.org/Vol-748/.

M. B. Caminati, M. Kerber, C. Lange, and C. Rowat.Proving soundness of combinatorial Vickrey auctionsand generating verified executable code. 2013. arXiv:1308.1779 [cs.GT].


http://ceur-ws.org/Vol-748/

http://arxiv.org/abs/1308.1779

References

References III

P. Cramton, Y. Shoham, and R. Steinberg, eds.Combinatorial auctions. MIT Press, 2006.

M. Kerber, C. Lange, and C. Rowat. ForMaRE. FormalMathematical Reasoning in Economics. url: http://cs.bham.ac.uk/research/projects/formare/(visited on 2013-02-10).




References

References IV

C. Lange, T. Mossakowski, O. Kutz, C. Galinski,M. Grüninger, and D. Couto Vale. “The DistributedOntology Language (DOL): Use Cases, Syntax, andExtensibility”. In: Terminology and KnowledgeEngineering Conference (TKE). (Madrid, Spain,June 20–21, 2012). Ed. by G. Aguado de Cea,M. C. Suárez-Figueroa, R. García-Castro, andE. Montiel-Ponsoda. 2012, pp. 33–48. arXiv:1208.0293 [cs.AI]. url: http://oeg-lia3.dia.fi.upm.es/tke2012/proceedings.


http://arxiv.org/abs/1208.0293

http://oeg-lia3.dia.fi.upm.es/tke2012/proceedings

http://oeg-lia3.dia.fi.upm.es/tke2012/proceedings

References

References V

C. Lange. “Towards OpenMath Content Dictionariesas Linked Data”. In: 23rd OpenMathWorkshop. Ed. byM. Kohlhase and C. Lange. July 2010. arXiv:1006.4057v1 [cs.DL]. url:http://cicm2010.cnam.fr/om/.

C. Lange. “Enabling Collaboration on SemiformalMathematical Knowledge by Semantic WebIntegration”. PhD thesis. Jacobs University Bremen,2011.


http://arxiv.org/abs/1006.4057v1

http://cicm2010.cnam.fr/om/

References

References VI

T. Mossakowski, O. Kutz, and C. Lange. “ThreeSemantics for the Core of the Distributed OntologyLanguage”. In: Formal Ontology in InformationSystems. 7th International Conference (FOIS 2012).(Graz, Austria, July 24–27, 2012). Ed. by M. Donnellyand G. Guizzardi. Frontiers in Artificial Intelligenceand Applications 239. (The paper has won the bestpaper award. Also published at IJCAI 2013 track on BestPapers in Sister Conferences.) Amsterdam: IOS Press,2012, pp. 337–352. url:http://interop.cim3.net/file/pub/OntoIOp/Publications/FOIS_2012/paper.pdf.


http://interop.cim3.net/file/pub/OntoIOp/Publications/FOIS_2012/paper.pdf

http://interop.cim3.net/file/pub/OntoIOp/Publications/FOIS_2012/paper.pdf

References

References VII

D. Vrandečić, C. Lange, M. Hausenblas, J. Bao, andL. Ding. “Semantics of Governmental Statistics Data”.In: Proceedings of WebSci’10: Extending the Frontiers ofSociety On-Line. Web Science Trust, 2010. url:http://journal.webscience.org/400/.


http://journal.webscience.org/400/

Linking Big Data to Rich Process Descriptions

Education

rdf data

data knowledge

structured data

introduction motivation

manufacturing process

formal verification

minute lange

semantic web integration