Web Ontologies: Lessons Learned from Conceptual Modeling at Scale

Professur für ALLGEMEINE BWL
insbesondere E-BUSINESS
Web Ontologies: Lessons Learned
from Conceptual Modeling at
Scale
Martin Hepp, @mfhepp
mfhepp@gmail.com

Ontology in Philosophy
•  “…that branch of philosophy which deals with the nature and the organisation
of reality.”
Guarino and Giaretta (1995):
Ontologies and Knowledge Bases
2

Example: Part-Whole Theories
• “Mereology”
• E.g. proper parts vs.
other forms of part-
whole relations
3

Sounds familiar?
4

Here is the full section:
Gruber, T. R. (1995):
Toward principles for the design of ontologies used for knowledge sharing.
5

Gruber (1995): Ontologies in Computer Science
The term is „borrowed from philosophy“
6

Ontologies Approximate the Intended Models
7
N. Guarino (1998):
Formal Ontology in Information Systems

Ontologies for Systems Interoperability
8
N. Guarino (1998):

Problem
9
N. Guarino (1998):

Ideal Ontology
N. Guarino (1998):
10

However,…
11
§  Computer-based Information Systems are not close to
be full knowledge-based systems
§  They are unable to operate solely on axiomatic
theories (“rule sets”)

This reduces the practical scope of ontologies to…
•  Providind a shared type system that is useful across systems
•  Improving the reliability of type information for entities
•  Providing some rules that hold implicit facts (e.g. transitivity of a property)
12

Improve the Reliability of Type Information
Type:Hotel
Hotel
1
Hotel
2
Hotel
3
System 1 System 2
13

Acceptable Ontology
Prof. Dr. Martin Hepp, Universität der Bundeswehr München14
Two users / systems

Bad Ontology
Two users / systems

Reliability of Type Membership (Sketch)
Some dimension
Agents
who agree
16

Historically, Ontologies aimed at deterministic behavior

The community was small

And there were social ties beyond HTTP between the humans
EKAW
1987
Researcher
1
Researcher
3
Researcher
2

And then came the Web

And people borrowed the term again...
Ontologies for the Web

The Semantic Web (2001)
Ontologies....
§  for increasing automated information
processing at Web scale

However, if the number of human users gets large,...

...the whole thing turns into a probabilistic system
§  Agents Ai, A set of all agents
§  Sender S element of A
§  Receiver R element of A
§  O set of objects / phenomena
§  f(Oi,Sj) = 0 or 1
§  f(Oi, Ri) = 0 or 1
§  Etc.

Reliability of Type Information
2 Agents, 85 % Agreement

The Fallacy of Raw Consumption of Web Data
http://www.heppresearch.com29
Naïve Type Membership Interpretation: SPARQL
# Find former people who are professors
PREFIX dbpedia-owl: <http://dbpedia.org/ontology/>
SELECT * {?s a dbpedia-owl:Professor} LIMIT 100

Naïve Type Membership Interpretation: SPARQL
Find all professors from Web markup
<html prefix="schema: http://schema.org/!
dbpedia: http://dbpedia.org/ontology/">!
!
<div typeOf="schema:Person dbpedia:Professor" about="#person">!
<span property="schema:honorificPrefix">Prof. Dr.</span> !
<span property="schema:givenName">Zaphod</span>!
<span property="schema:familyname">Beeblebrox</span>!
</div>!
</html>

Which properties determine the probabilities?
§  Number of individuals of that type
–  Familiarity: Banana vs. Papaya
§  Cognitive skills of involved humans
§  Etc.
§  We do not know yet ;-)

Fallacy 1: Formal specifications would guarantee correct
usage
§  Axiomatic definitions
§  In fact, no serious evidence

Fallacy 2: Web Ontologies allow for the automated processing
of data
§  Raw data on the Web can rarely be used directly for
any meaningful purposes
§  Data cleansing, entity consolidation, etc.
§  Navigation on raw, linked data questionable

Recommended Reading

The Role of Schemas between Men and Machines
DatabaseSchemas
WebOntologies

Test: How well can the target audience understand and apply
the conceptual distinctions in the ontology?
§  Examples:
–  Book copy vs. book title
–  Web page vs. entity described by the page
–  Legal entity vs. shop location
§  Turning ontology engineering from an art into an
empirical science.

Optimize: Compare the discriminative value of your
conceptual elements with their effect on conformant ontology
usage

Monitor: Capture live feedback on the quality of the
membership functions of your ontology

Challenges
•  Goal 1: Improve the reliability of type memberships of
entities in information processing
•  Goal 2: Find types that provide sufficient distinctions for an
algorithmic information processing, i,e. minimize
reclassification tasks at the recipient’s side
•  Goal 3: Ditto for data granularity
•  Goal 4: Find types that can be populated from existing
datasources without reclassification tasks at the publisher
side.
•  Goal 5: Find conceptual distinctions that can be reliable
mastered by the human stakeholders of the systems
42

Challenges
side.
43

Challenges
side.
44

Challenges
side.
45

Challenges
side.
46

What is schema.org? What is GoodRelations?
1. Official Characterization
2. Purpose:
§  Focus on information extraction on the Web
§  Other uses as a by-product
3. Knowledge Representation Perspective
§  Entity Types
§  Relationship Types
§  Weak Domain / Range Semantics
§  Syntax-independent Meta-Model
And how are they related?
Questions? Suggestions? Contact me at @mfhepp!47

Official Characterization from http://schema.org
This site provides a collection of schemas that webmasters can use
to markup HTML pages in ways recognized by major
search providers, and that can also be used for
structured data interoperability (e.g. in JSON). Search
engines including Bing, Google, Yahoo! and Yandex rely on this markup to improve the
display of search results, making it easier for people to find the right Web pages.
Many sites are generated from structured data, which is often stored in
databases. When this data is formatted into HTML, it becomes very difficult to
recover the original structured data. Many applications, especially
search engines, can benefit greatly from direct access to this structured
data. On-page markup enables search engines to understand the information on web
pages and provide richer search results in order to make it easier for users to find
relevant information on the web. Markup can also enable new tools
and applications that make use of the structure.

There is REAL Momentum
A lot of data
§  Since 2011, schema.org has been added to >25% of top-ranked e-
commerce sites product detail pages.
§  RDF-based representations are specified.
Table: Random sample of n=73 product detail pages from high-ranking Google results.
Note that these numbers have a strong bias towards popular, professionally operated sites.

Schema.org: A Data Publication Ontology
Not designed for raw data consumption (only as a by-product)
§  Historically, ontologies in computer
science aimed at harmonizing the
conceptualization and representation
of data for publishers and consumers
of the data.
§  Implicit goal of the traditional Semantic
Web stack: More or less, consumption
of raw data.
§  This requires detailed consensus on
the level of data granularity and data
semantics at scale, and high data
quality.
§  Schema.org does not make this
assumption, since its sponsors have
the power to work on semi-structured
data at Web scale.
Ontologyschema.org

Schema.org: The Semantic Web Vision Come True?
1. No OWL. Not even an ontology in the narrow sense.
2. Direct consumption difficult
§  Crawling
§  Cleansing
§  Lifting
3. No broad use of Linked Data principles
§  Mostly no global entity identifiers
§  Page = Entity (vs. httpRange-14)
§  No vocabulary reuse (*)
Likely not what the Semantic Web community had hoped for.

From Ontology to Ontologies: The First Passage
•  From Philosophy to Computer Science
•  Exchange of Knowledge Bases
52

The Web as an Ecosystem for Shared Schemata
•  Stakeholders
–  Schema Designer (a few)
–  Data Consumer (set the incentives)
–  Data Publisher / Site Owner (many, often limited set of skills and
understanding)
53

Web Ontology ≠ Ontology
•  Huge, unknown community of users
–  You speak only via your Web resources to them
•  Economic factors
–  Incentives vs. Costs, Spam, …
•  Data Quality issues
•  Huge amount of data
•  Limitations of deployment environments
–  Access to server configuration
–  Skills
–  Corporate policies
54

The challenges on the Web are the same.
Just at a bigger scale.
side.
55

Ontology Perspicuity
How well can the ontology be understood through the imperfect channel of
communication between the designers and the users of an ontology?
56

Ontology Economics
57

Maintenance and Dynamics
58

Requirements on Web Ontologies
•  Broad user base (>500)
•  Understandable and usable just from the Web presence
•  Decoupling the ontology from the evolution of enumerations
•  Dynamic Data Granularity and Incremental Refinement
•  Deferred Consensus
•  HTML documentation
•  Proper deployment
•  Extensions should work with clients that understand only the basics if they have
minimal RDFS reasoning support
•  Minimize namespace traversals, i.e., locally define most concepts and link to
external ontologies than simply reuse
•  Minimize number of conceptual elements
59

Types of Community Involvement
•  Centralized Ontology Engineering
•  Community-driven Ontology Engineering
•  Community-inspired Ontology Engineering
•  Social Functionality (as in GoodRelations)
60

Web Ontology Techniques
•  Dynamic Degree of Disambiguation
•  Dynamic Data Granularity and Lexical Carry-Over
•  Deferred Consensus
•  Incremental Refinement
•  Optional Nodes for N-Ary Relations
•  Social Functionality
•  Externalize Enumerations
61
(Hepp 2015, forthcoming)

Dynamic Degree of Disambiguation
ProductOrService
Service Product
Agent
Person Organisation
62

Dynamic Data Granularity
63

Example for Deferred Consensus
gr:name
gr:description
schema:geo schema:GeoCoordinates
acco:size gr:QuantitativeValue
acco:occupancy gr:QuantitativeValue
acco:occupancyAdults gr:QuantitativeValue
acco:occupancyInfants gr:QuantitativeValue
acco:petsAllowed xsd:boolean
acco:Accommodation
acco:feature acco:propertyID (*) xsd:string
gr:name (for raw values/text)
gr:description
acco:min (gr:hasMinValue) gr:QuantitativeValue
acco:max (gr:hasMaxValue) gr:QuantitativeValue
acco:value (gr:hasValue) gr:QuantitativeValue
acco:unit text
acco:unitCode xsd:string (gr:hasUnitOfMeasurement)
acco:AccommodationFeature
PropertyIDs
starRating:HOTREC
starRating:DEHOGA
starRating:WHR
starRating:Hotelstars
64

GoodRelations: Language Reference
http://purl.org/goodrelations/v1.html
Examples
Links and
References
Range and Domain
Social MediaCommunity
of Experts
65

UML Class Diagram
http://www.heppnetz.de/ontologies/goodrelations/v1#uml
The GoodRelations Ontology for E-Commerce
Language Overview - UML Class Diagram
http://purl.org/goodrelations/
Version 1, Release 2011-10-01
Martin Hepp. mhepp@computer.org
gr:name rdfs:Literal
gr:description rdfs:Literal
gr:hasEAN_UCC-13 xsd:string
gr:hasGTIN-14 xsd:string
gr:hasStockKeepingUnit xsd:string
gr:hasMPN xsd:string
gr:condition rdfs:Literal
gr:serialNumber xsd:string
gr:eligibleRegions xsd:string
gr:validFrom xsd:dateTime
gr:validThrough xsd:dateTime
gr:availabilityStarts xsd:dateTime
gr:availabilityEnds xsd:dateTime
gr:category rdfs:Literal
gr:Offering
gr:datatypeProductOrServiceProperty any
gr:hasEAN_UCC-13 xsd:string
gr::hasGTIN-8 xsd:string
gr:hasStockKeepingUnit xsd:string
gr:hasMPN xsd:string
gr:condition rdfs:Literal
gr:color rdfs:Literal
gr:ProductOrService
gr:priceType xsd:string
gr:hasUnitOfMeasurement xsd:string
gr:billingIncrement xsd:float
gr:UnitPriceSpecification
gr:DeliveryChargeSpecification
gr:legalName rdfs:Literal
gr:hasNAICS xsd:int
gr:hasISICv4 xsd:int
gr:hasDUNS xsd:string
gr:hasGlobalLocationNumber xsd:string
gr:BusinessEntity
gr:WarrantyScope
gr:PaymentMethod
gr:ProductOrServiceModel
gr:hasMinValue rdfs:Literal
gr:hasMaxValue rdfs:Literal
gr:hasValue rdfs:Literal
gr:QuantitativeValue
gr:durationOfWarrantyInMonths xsd:int
gr:WarrantyPromise
gr:DayOfWeek
gr:DeliveryMethod
gr:QualitativeValue
gr:amountOfThisGood xsd:float
gr:TypeAndQuantityNode
gr:BusinessFunction
gr:hasGlobalLocationNumber xsd:string
gr:hasISICv4 xsd:int
gr:Location
gr:PaymentMethodCreditCard
gr:BusinessEntityType
gr:hasMaxValueInteger xsd:int
gr:hasMinValueInteger xsd:int
gr:hasValueInteger xsd:int
gr:QuantitativeValueInteger
gr:SomeItems
gr:opens xsd:time
gr:closes xsd:time
gr:OpeningHoursSpecification
gr:PaymentChargeSpecification
gr:DeliveryModeParcelService
gr:hasCurrency xsd:string
gr:valueAddedTaxIncluded xsd:boolean
gr:hasMaxCurrencyValue xsd:float
gr:hasMinCurrencyValue xsd:float
gr:hasCurrencyValue xsd:float
gr:PriceSpecification
gr:hasMinValueFloat xsd:float
gr:hasMaxValueFloat xsd:float
gr:hasValueFloat xsd:float
gr:QuantitativeValueFloat
gr:seeks gr:includesObject
gr:typeOfGood
gr:hasMakeAndModel
gr:hasMakeAndModel
gr:hasManufacturer
gr:hasPOS
gr:hasOpeningHoursSpecification
gr:availableAtOrFrom
gr:hasEligibleQuantity
gr:hasWarrantyPromise
gr:appliesToPaymentMethod
gr:hasWarrantyScope
gr:qualitativeProductOrServiceProperty
gr:quantitativeProductOrServiceProperty
gr:appliesToDeliveryMethod
gr:hasBusinessFunction
gr:availableDeliveryMethods
gr:eligibleCustomerTypes
gr:acceptedPaymentMethods
gr:hasPriceSpecification
gr:hasOpeningHoursDayOfWeek
gr:isConsumableFor
gr:isVariantOf
gr:isSimilarTo
gr:isAccessoryOrSparePartFor
gr:offers
gr:hasInventoryLevel
gr:typeOfGoodgr:includes
gr:includes
Notes:
1. The following GoodRelations elements are only shortcuts for simpler annotation or querying. See the documentation at http://purl.org/goodrelations/ for details:
gr:hasValue (shortcut for setting both hasMinValue and hasMaxValue properties to the same value in one turn)
gr:hasValueFloat (shortcut for setting both hasMinValueFloat and hasMaxValueFloat properties to the same value in one turn)
gr:hasValueInteger (shortcut for setting both hasMinValueInteger and hasMaxValueInteger properties to the same value in one turn)
2. The following elements are now deprecated, but you can still use them, e.g. for staying compatible with older data consumers (e.g. Yahoo SearchMonkey):
gr:ActualProductOrServiceInstance (now gr:Individual)
gr:ProductOrServicesSomeInstancesPlaceholder (now gr:SomeItems)
gr:LocationOfSalesOrServiceProvisioning (now gr:Location
3. For the recommended cardinality of attributes, see the GoodRelations Language Reference at http://purl.org/goodrelations/ v1.html.
4. gr:valueReference links can also exist between a gr:QualitativeValue and a gr:QuantitativeValue and vice versa, but this rare case is not shown for readability.
5. gr:name and gr:description can now be attached to any GoodRelations type, but this is not shown here for readability.
gr:eligibleDuration
gr:advanceBookingRequirement
gr:owns
gr:greater
gr:lesser
gr:lesserOrEqual
gr:greaterOrEqual
gr:equal
gr:nonEqual
gr:hasPrevious
gr:hasNext
gr:successorOf
gr:predecessorOf
gr:eligibleTransactionVolume
gr:eligibleTransactionVolume
gr:deliveryLeadTime
gr:serialNumber xsd:string
gr:Individual
Red highlighting indicates elements added or changed in this release.
gr:valueReference
gr:valueReference
gr:hasEligibleQuantity
gr:includes
gr:hasInventoryLevel
gr:weight
gr:width
gr:height
gr:depth
gr:Brand
gr:hasBrand
gr:hasBrand
gr:hasBusinessFunction
gr:License
gr:eligibleDuration
66

GoodRelations User‘s Guide
http://wiki.goodrelations-vocabulary.org/Documentation
Will be updated soon!
67

Community of Experts
68

Albert Einstein on Ontology Engineering ☺
"Make everything as simple as possible, but not simpler.“
Albert Einstein
69

Let’s Do Science, not Cult!
§  Challenge paradigms and
approaches
§  Use hard data, not beliefs and
assumptions (neither your own ones
nor the ones inherited from the old
folks)
CC BY-SA 3.0 / Nicor / https://en.wikipedia.org/wiki/North_Korea's_cult_of_personality#/
media/File:Mansudae_Grand_Monument_08.JPG

Thank you!
@mfhepp
71

Web Ontologies: Lessons Learned from Conceptual Modeling at Scale

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Was ist angesagt?

Was ist angesagt? (12)

Ähnlich wie Web Ontologies: Lessons Learned from Conceptual Modeling at Scale

Ähnlich wie Web Ontologies: Lessons Learned from Conceptual Modeling at Scale (20)

Mehr von Martin Hepp

Mehr von Martin Hepp (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Web Ontologies: Lessons Learned from Conceptual Modeling at Scale