Ontology Engineering: A View from the Trenches - WOP 2015 Keynote

The Big Picture Challenges Architecture Patterns Semantic Signatures Interfaces
Ontology Engineering:
A View from the Trenches
WOP 2015 Keynote
Krzysztof Janowicz
STKO Lab, University of California, Santa Barbara, USA
Ontology Engineering: A View from the Trenches K. Janowicz

The Big Picture
The Big Picture∗

The Big Picture
The Even Bigger Picture
Ontology Engineers
Oscar Corcho’s EKAW2014 Keynote: Ontology engineering for and by the
masses: are we already there? (http://goo.gl/loYAta)
Ontology
The next 60+ minutes
Ontology Engineering
Pascal Hitzler’s Diversity++ 2015 Keynote: Ontology modeling with
domain experts (see.it/tomorrow/9am)

Ontology Engineering Challenges
Ontology Engineering
Challengesby Example

Ontology Reuse
Vocabulary/Ontology Reuse
A typical statement:
’Reuse external vocabularies whenever possible.’
<http://dbpedia.org/resource/Copernicus_(lunar_crater)>
...
geo:lat
"9.7"^^xsd:decimal;
geo:long
"20.0"^^xsd:decimal;
...
a
dbpedia-owl:Crater,
...
ns5:Place,
...
Concerns:
Most ontologies are under-speciﬁc, the intended meaning or scope remains
unclear, versioning /evolution strategies are unclear, contact persons are
often not available, potential legal issues, lack of proper documentation,
diﬀerent community-based approaches and styles,...

Ontology Reuse
Reuse Difficulties Example
The Fluidops interface renders the DBpedia RDF data from the Copernicus
crater and places it on the surface of the Earth instead of realizing that the
given coordinates are selenographic coordinates.

Standardizing Meaning
‘Standardized Meaning’ Approaches to Ontology Engineering

Standardizing Meaning
‘Standardized Meaning’ Approaches to Ontology Engineering
California:
City ≡ Town
Utah:
Town ≡ < (population, 1000)
Pennsylvania:
Town ≡ {Bloomsburg}
We will revisit the cities & towns example and the local nature of meaning later.

Styles and Committments
Modeling Practive, Styles, and Level of Detail
With growing size, complexity, and abstraction-level, diﬀerent modeling styles,
ontological choices, levels of detail becomes increasingly diﬃcult to handle.

The Cause of the Problem
Is There a Common Cause to These Problems?

The Cause of the Problem
Is There a Common Cause to These Problems?
We will revisit the age example at the end of this talk.

Architecturefrom above

Pattern-based Arcitecture
Envisioned, Pattern-based Arcitecture (Horizontal)
Patterns act as fallback level that ensures minimal interoperability while
preserving heterogeneity (i.e., local, repository-speciﬁc ontologies can diﬀer).

Views as virtual ontologies. ‘All’ provider- and user-perspectives agree on a
common core; more speciﬁc results can diﬀer.

All users can query for data that correspond to the pattern, using view A one can
retrieve data on human trajectories but these data will only come from RA and RB.

Envisioned, Pattern-based Arcitecture (Vertical)

Envisioned, Pattern-based Arcitecture (Vertical)
Cons
... speak now or forever
hold your peace
Pros
Defers the introduction of classes that are
heavy on ontological commitments (e.g.,
‘vulnerability’)
No need for (community-wide) agreement
which is a key (social) challenge for other
approaches. Preserves heterogeneity
Mines ontological primitives out of real
observation data
Assists domain experts in becoming
knowledge engineers by developing
reusable patterns
Moves ontology reuse to the layer where
it belongs (and avoid Frankenontologies)
Is driven by publishing, discovery, reuse,
and integration needs.

Patternsand views

Semantic Trajectory Pattern
A Semantic Trajectory Pattern
A pattern for discrete trajectories of people, wildlife, vessels, and so forth.

A Semantic Trajectory Pattern

Ontology Design Patterns Can Be Specialized
Trajectories that model the research cruises of scientiﬁc vessels
Is this still an ontology design pattern?

A Micro-Ontology for Cruises
Combining the InformationObject, Agent, Event, Vessel, and Trajectory patterns

Idea Behind Semantic Trajectory Pattern
Can cover a wide range of domains
Can be easily extended
Supports multiple granularities
Axiomatization beyond mere surface semantics
Has various hooks to well-known ontologies / patterns.
Only partially self-contained

Typecasting and Views
Typecasting – Dealing with Styles
Typecasting Individual to Class and Back
ClassName ∃hasType.{classname} (1)
∃hasType.{classname} ClassName (2)
Roliﬁcation: Typecasting from Classes to Properties
...
Reiﬁcation: Typecasting Properties into Classes
...

Views – Dealing with Granularity and Perspectives
Vessel ≡ ∃RVessel.Self
Cruise ≡ ∃RCruise.Self
Trajectory ≡ ∃RTrajectory.Self
Segment ≡ ∃RSegment.Self
RSegment ◦ hasSegment−
◦ RTrajectory ◦
◦ hasTrajectory−
◦ RCruise ◦ isUndertakenBy isTraversedBy

Signatures
Semantic

Sensors and Signatures
Observatories and Their Sensors
Whether on land or in space, observatories and their sensors serve
diﬀerent purposes and are most useful when they work together.

Sensors and Signatures
Spectral Signatures, Bands, and Remote Sensing
Spectral signatures are the combination of emitted, reﬂected, or absorbed
electromagnetic radiation at varying wavelengths (bands) that uniquely
identify a feature type.
Spectral libraries, the idea of sharing spectral signatures, has
revolutionized remote sensing.

Semantic Signatures and Bands
Semantic Signatures As Analogy To Spectral Signatures
Geospatial bands
based on geographic location
ANND
Ripley’s K Bins
J Measure
Dzero
Temporal bands
based on geo-social check-ins
24 Hours
7 Days
Seasons
Thematic bands
based on venue tips and reviews
LDA topics
TF-IDF
Makes use of data
heterogeneity

Semantic
signature
Geographic
feature
Feature type
Spatial band
Temporal band
Thematic band
rdfs:subClassOf
rdfs:subClassOf
rdfs:subClassOf
signiﬁes
hasObservationResulthasType
BandconsistsOf
Spatial signature
Temporal
signature
Thematic
signature
rdfs:subClassOf rdfs:subClassOf rdfs:subClassOf
LDA topic vector
rdf:type
hasSignature
Semantic signatures and bands are an analogy to spectral signatures.
So far, we have mined and modeled hundreds of bands for hundreds of
diﬀerent geographic features on the micro, meso, and macro-scale.
Applied them to categorization, deduplication, semantic enrichment, cleansing,
visualization, exploration, reverse geocoding, ontology alignment,...

Thematic Bands
Thematic Bands & Geo-Indicativeness
Places at geographic location 34.43, -119.71 are:
of types city, county seat,...
at the coastline, near the mountains, have Mediterranean climate,...
described in terms of urban area, economy, tourism, government, employment,...
Interesting observation: some of these terms will co-occur by type, others per region.

Thematic Bands
Thematic Bands & Geo-Indicativeness
A thematic band can be
computed out of unstructured
text from sources such as
Wikipedia, travel blogs, news
articles, and so forth.
Non-georeferenced plain text
is often still geo-indicative
Diﬀerent types of geographic
features have diﬀerent,
diagnostic topics associated to
them (out of 500 topics)
Indicative topics and be lifted to
the type-level.
Here, we modeled topics using
latent Dirichlet allocation (LDA)

Thematic Bands
Thematic Bands & Geographic Feature Types
City topics: 204>450>104>282>267>497>443>484>277>97>...
Town topics: 425>450>419>367>104>429>266>69>204>308>...
Mountain topics: 27>110>5>172>208>459>232>398>453>183>...

Temporal Bands
Temporal Bands
Study geo-social
check-in data to
location-based social
networks.
Aggregate them to the
feature type level and
clean them.
Intuitively, people visit
wineries in the
after-noon and evening
and bakeries in the
mornings.
Combining weekly and
hourly bands to create
place type signatures.

Spatial Bands
Spatial Bands
POI plotted by similarity to bar and post oﬃce in OpenStreetMap data (London)
Similarity measured as association strength in OSM change history
Bars (and similar features) tend to clump together
Post Oﬃces (and similar features) are rather uniformly distributed

Spatial Bands
Spatial Bands
Dzero measures the likelihood of features of a certain type to co-occur
within a speciﬁc semantic and spatial range.
General idea: generate recommendations and clean up data based on
type likelihood. ’How likely is a post oﬃce directly next to an existing one?’

Sensor Resolution & Social Sensing
(Remote sensing) sensors can be characterized by their resolution
Spatial resolution: smallest feature that can be detected, i.e., the pixel size.
Temporal resolution: smallest time interval between a repeated observation.
Spectral resolution: number, position, and width of spectral bands.
Radiometric resolution: small distinguishable diﬀerences in radiation magnitude.
Analogous social sensor resolutions, e.g., types of bands, number of topics.

Place-based Resolution of Termporal Signatures
Circular temporal signatures histograms for Theme Park (a,b,c) and Drugstore (d,e,f).
About 50% of ≈ 400 Point Of Interest (POI) types are regionally invariant in the USA.

Temporal Resolution of Termporal Signatures
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
The ‘Foursquare-day’
How and when do people check-in at places, manually, automatically?
Do they check-out? If not, after what time are they checked-out automatically?

Distinguishable Feature Types For Thematic Signatures From 500-Topics
Which place types can be meaningfully distinguished (in DBpedia)?

Interfaces
Ontologies as

Exploring Linked Data
Ontologies as Interfaces
Our completely client-based JS explorer can be connected to any triple store

Allows users to explore Linked Data; constructs ﬁlters (e.g., >) based on probing

How does it know which data can be mapped and how to do this?

Our explorer can even combine data from multiple sources as layers (here cruises
from R2R in green and gazetteer features in pink)

What about Time?
What about Time and Age?
Google’s Knowledge Graph in 2012

What about Time?

What about Time?
Will you please deﬁne a rule how age is computed... (like a Time Interface )

Ontology Engineering: A View from the Trenches - WOP 2015 Keynote

Recommended

Recommended

More Related Content

Similar to Ontology Engineering: A View from the Trenches - WOP 2015 Keynote

Similar to Ontology Engineering: A View from the Trenches - WOP 2015 Keynote (20)

More from kjanowicz

More from kjanowicz (15)

Recently uploaded

Recently uploaded (20)

Ontology Engineering: A View from the Trenches - WOP 2015 Keynote