An Ontology-based Decision Support Framework for Personalized Quality of Life Recommendations

An Ontology-Based Decision Support Framework
for Personalized Quality of Life Recommendations
Marina Riga1,2, Efstratios Kontopoulos1, Kostas Karatzas2, Stefanos Vrochidis1,
Ioannis Kompatsiaris1
1CERTH-ITI, Information Technologies Institute, Thessaloniki, Greece
2ISAG-EI Group, Dept. of Mech. Engineering, Aristotle University of Thessaloniki, Greece
4th International Conference on Decision Support System Technology –
ICDSST 2018 & PROMETHEE DAYS 2018
22 – 25 May 2018, Heraklion, Crete, Greece

Outline
2
Introduction
 Focus, aim and theoretical background
The proposed ontology-based approach
 Domain’s knowledge modelling
 Reasoning
 Adoption
Advantages
Example use case
Evaluation
Conclusions & Future work

Focus
3
Domain:
 Air Quality (AQ)
 Decision support systems (DSS)
− Information: location, incident, individual
− Personalized aspect
Issues:
 heterogeneity
 legal framework
 scale
Use: ontologies

So, what about ontologies?
4
Definition: explicit specification of a domain of
discourse, in a structured and semantically rich way
[Gruber, 1993]1
OWL 2 Web Ontology Language
 classes, properties, individuals
 triple <?subject ?predicate ?object>
Advantages
 semantic expressiveness
 understanding
 sharing
 reuse of data
1 Gruber, T.R.: A translation approach to portable ontology specifications. Knowledge acquisition J. 5(2), pp. 199-220 (1993)

Ontologies in the environmental domain
5
Abstract or domain specific
 SWEET: physical/ecological phenomena, meteorology, etc.
 PESCaDO: personalized information and decision support
 EnvO: habitats, materials, etc.
 AIR_POLLUTION_Onto: AQ analysis and control
Different tasks:
 Semantic declaration and easy access
 Integration, linking of data and heterogeneous content
 Multi-criteria selection

The proposed framework (1/2)
6
Handles both the static and the dynamic processes of a
DSS operation
1. A set of ontological concepts and relations for
representing definitions of:
 User profile characteristics (age, preferred activities, …)
 Health sensitivities (sensitive groups) and risks
 Geo-related AQ measurements
 Requests
 Personalized recommendations
2. Novel rule-based reasoning mechanism to support
personalized recommendation provision

The proposed framework (2/2)
7
Three basic components of a general DSS:
i. the data
ii. the model
iii. user-system interaction

I. Domain knowledge modelling
8
Documentation and sources:
http://mklab.iti.gr/project/hackair-ontologies

II. Reasoning over domain knowledge
9
SPIN rules – semantic interpretation and inference
Two-layered implementation
1. 109 rules: schema + instances → low-level derivations
• age values → age groups
• AQ observations → AQI scales
• person → defined user profile(s)
2. 150 rules: schema + preceding inferences → higher-level
deductions

III. Communication with ontology-based
framework
10
RESTful web services (POST request, JSON input)
 Dynamic population of new instances
 Ontology-based problem description
 Triggering of reasoning mechanism
Tools: Java EE7, Apache Jena, SPIN API
Operational use: hackAIR platform & app
(http://www.hackair.eu/)

Use case
11
Valeria, a 32-year-old woman, pregnant with respiratory problems (asthma),
usually goes for long walks in the city. She queries the hackAIR application
for information about existing AQ conditions. At the time of request, the
PM10 values are extremely high (e.g. 150μg/m3)
Defined triples Inferred triples
abox:Valeria
rdf:type TBox:Person ;
tbox:hasAge 32^^xsd:integer ;
tbox:hasGender tbox:female ;
tbox:isSensitiveTo tbox:Asthma ;
tbox:isPregnant "true"^^xsd:boolean ;
tbox:hasLocation abox:location_V ;
tbox:hasPreferredActivity tbox:walking_activity ; .
abox:location_V
tbox:hasEnvironmentalData abox:PM10EnvData_V .
…
…
abox:Valeria
rdf:type tbox:AdultPerson ;
rdf:type tbox:PregnantFemalePerson ;
rdf:type tbox:SensitiveHealthPerson ;
rdf:type tbox:SportsWalkingPerson ;
rdf:type tbox:Pregnant_Sensitive_Person ;
rdf:type tbox:Pregnant_SportsWalking_Person ; .
abox:location_V tbox:hasRelatedIndex tbox:AQI_bad ; .
abox:Valeria
tbox:isProvidedWithRecommendation [
rdf:type tbox:LimitExposureRecommendation ;
tbox:hasDescription "You should go for a walk in an area with cleaner air."@en ;
tbox:hasDescriptionIdentifier "walking" ; ] ;
tbox:isProvidedWithRecommendation [
rdf:type tbox:LimitExposureRecommendation ;
tbox:hasDescription "Consider avoiding any intense outdoor activity in your area! The existing
air quality might be harmful for your health."@en ;
tbox:hasDescriptionIdentifier "general personalised" ; ] ; .

Evaluation
13
Consistency of the provided results
 Checked by ontology-experts
Performance of the overall service
 11 different problems: simple + complex profiles
 Response time: 1.32 – 1.89s
Quality of the proposed solution
 lightweight, fast computations
User evaluation
 hackAIR pilots are still ongoing, thus results are
pending..

Key advantages
14
Uniformity
 SPIN rules compliant to ontology definitions
Flexibility
 multilingual definition
 “bag” of recommendations
Modularity
 rules’ content
 execution order
Extensibility
 support additional pollutants, sensitivities

Conclusions
15
Novel ontology-based DS framework
 Static & dynamic processes
 Uniformly structured & semantically enriched DS
knowledge base
Three-layered structure
1. Ontological schema
2. Structured actual data, with semantic extension
3. Rule-based reasoning
Achievements
 Fast computations
 Modularity + extensibility of the system
 Domain agnostic

Future work
16
User- and/or AQ expert- centered evaluation
Fuzzy reasoning
 Represent more complex relations
Ontology Design Pattern (ODP)
 Propose modelling solutions for main building blocks
SHACL framework – SPIN’s next generation

Thank you!
mriga@iti.gr
mriga@isag.meng.auth.gr
Credits
hackAIR project (http://www.hackair.eu/)
“Collective awareness platform for outdoor air pollution”
This work has received funding from the EU Horizon 2020 Research and
Innovation Programme (GA 688363)
17

An Ontology-based Decision Support Framework for Personalized Quality of Life Recommendations

More Related Content

Similar to An Ontology-based Decision Support Framework for Personalized Quality of Life Recommendations (20)

Recently uploaded (20)

An Ontology-based Decision Support Framework for Personalized Quality of Life Recommendations

Editor's Notes