Shiyan Ou1, Viktor Pekar1, Constantin Orasan1,

1
Development and Alignment of a Domain-Specific
Ontology for Question Answering

• Shiyan Ou1, Viktor Pekar1, Constantin Orasan1,
• Christian Spurk2, Matteo Negri3
• 1Research Group in Computational Linguistics,
  University of Wolverhampton, UK
• 2German Research Centre for Artificial
  Intelligence GmbH (DFKI), Germany
• 3Fondazione Bruno Kessler FBK, Italy

2
Structure

• Introduction to QALL-ME
• The QALL-ME ontology
• Alignment to WordNet and SUMO
• How the ontology is used for data encoding
• Conclusions

3
Introduction to QALL-ME

• QALL-ME (Question Answering Learning technologies
  Multilingual Multimodal Environment) is an
  EU-funded project which aims to establish a
  shared infrastructure for multilingual and
  multimodal question answering in the domain
  tourism.
• Projects website http//qallme.fbk.eu/
• In the QALL-ME system
• users pose natural language questions in several
  languages (both in textual and speech modality)
  using a variety of input devices (e.g. mobile
  phones), and
• returns a list of specific answers formatted in
  the most appropriate modality, ranging from small
  texts, maps, videos, and pictures.
• A domain-specific ontology for the tourism domain
  was developed and shared among all the partners.

4
The ontology in the project
WP3 Multilingual question interpretation
WP4 Annotation of entities Indexing of
data Retrieval of data
QALL-ME ontology
WP5 Multilingual answer extraction
WP9 Evaluation
See more in O39 Multilingual Resources
(Ambasadeurs) at 1305
5
Design of the ontology

• Analysis of data from content providers
• Analysis of users requirements
• Inspired by similar ontologies such as Harmonise,
  eTourism, Hi-Touch, TAGA, GETESS
• Harmonise and eTourism focus on static
  information (e.g. accommodation and
  events/activities), rather than dynamic
  information related to travel business (e.g.
  customers and itineraries) as the TAGA and
  Hi-Touch ontologies do.
• Similar to eTourism as is written in OWL rather
  RDFs
• but wider coverage than each individual existing
  ontology
• Introspection

6
Technical details of the ontology

• Encoded using OWL DL, since it has more
  expressive power than OWL Lite and has more
  efficient reasoning support than OWL Full
• Used Protégé-OWL as the editor and RacerPro7 as
  the reasoner
• The ontology contains
• 122 classes (concepts),
• 55 datatype properties and
• 52 object properties which indicate the
  relationships among the 122 classes.
• 15 top-level classes.
• The class hierarchy has a maximum depth of 4.

7
Part of the ontology (cinema/movies)
8
Ontology alignment

• The QALL-ME ontology was designed as a model of
  the narrow knowledge domain of tourism.
• The QALL-ME ontology was complemented with
  information from WordNet (and implicitly
  MultiWordNet) and SUMO via alignment
• The QALL-ME ontology is being changed so fully
  manual alignment was not a solution
• Fully automatic alignment is not precise enough,
  but maybe semi-automatic alignment is a solution

9
Ontology alignment (II)

• The alignment relied on
• String similarity of element identifiers (e.g.
  chalet ? chalet_1, SiteFacilityForChildren ?
  facility_)
• Structural similarity for disambiguation (e.g.
  uses the semantic distance to aligned concepts)
• Definition similarity for disambiguation
  (similarity between comments in the ontology and
  WordNet glosses is used)
• Structural similarity for unmatched concepts is
  calculated to all the nouns in WordNet

10
Ontology alignment (III)

• The overall accuracy of the fully automatic
  alignment is clearly suboptimal (precision of 49
  and recall of 31),
• Error analysis
• We noticed that for concept names with
  unambiguous matches in WordNet the algorithm
  performs without any errors
• The poor disambiguation performance is due to the
  very different depths of the two ontologies
• Only a few concepts have comments which are
  useful for definition similarity
• Semi-automatic alignment requires under 30
  minutes to obtain perfect alignment

11
Example of alignment

• QALL-ME SUMO WN2.1 WN2.1 gloss
• Accommodation _at_inhabits 02647858 living
  quarters provided for
• public convenience
• "overnight accommodations
• are available"
• Chalet _at_Building 02973228 a Swiss house with
  a sloping
• roof and wide eaves or a
• house built in this style
• PostOffice _at_Organization 08034771 an
  independent agency of the
• federal government
• responsible for mail delivery

12
Semantic annotation and database organization

• The ontology was used to encode the data
• Annotated data from the content providers was
  converted to RDF triplets
• The RDF documents can be stored in databases or
  plain text files
• The Jena RDF API was used for the operations

13
Semantic annotation and database organization
14
Content retrieval

• For retrieval SPARQL is used
• SPARQL is a query language for accessing RDF
  graphs by the W3C RDF Data Access Working Group
• SPARQL provides interoperability between languages

15

• What movie starring Halle Berry is on in
  Birmingham?
• Class MovieShow ?
• Property isInSite, Range Cinema ?
• Property hasPostalAddress, Range
  PostalAddress ?
• Property isInDestination, Range Destination
• Property name, Range string ltBirminghamgt
• Property hasEventContent, Range Movie ?
• Property name, Range string ltunknowngt
• Property hasStar, Range Star ?
• Propertyname, Range string ltHalle Berrygt

16

• PREFIX qme http//qallme.itc.it/ontology/qallme-t
  ourism.owl
• PREFIX xsd http//www.w3.org/2001/XMLSchema
• SELECT ?movieName
• WHERE
• ?MovieShow qmeisInSite ?Cinema.
• ?Cinema qmehasPostalAddress ?PostalAddress.
• ?PostalAddress qmeisInDestination ?Destination.
• ?Destination qmename Birminghamltxsdstringgt
• ?MovieShow qmehasEventContent ?Movie.
• ?Movie qmename ?movieName.
• ?Movie qmehasStar ?Star.
• ?Star qmename Halle Berryltxsdstringgt

17
Conclusions

• The QALL-ME ontology was specifically designed
  for the domain of tourism
• The ontology is playing an important role in
  several parts of the project
• The current ontology went through several
  revisions before reaching the current stage (and
  it may change again!!!)
• Both the ontology and its alignment to WordNet
  and SUMO will be made freely available on the
  projects website

18
Thank you !
Projects website http//qallme.fbk.eu/