Knowledge Graphs for Social Good: three challenges
Victor de Boer
With input from: Anna Bon, Christophe Guéret, Stephane Boyera , Nana Baah Gyan, Chris van Aart, Max Froumentin, Aman Grewal, Mary Allen, Amadou Tangara, Etienne Barnard, Hans Akkermans, André Baart, Gossa Lo, Myrthe van der Wekken, Onno Valkering, Fahad Ali, Romy Blankendaal, Stefan Schlobach, the Pressing Matter team…
four
Knowledge Graphs and the Semantic Web
Knowledge Graphs for Social Good
KGs for the rural poor: three challenges
A fourth challenge: polyvocality
Infrastructure
Interface
Relevancy
More and more structured data available online
Government data
Social web data
Medical data
Museum data
Research data
Development data
Moverum.com
CC-by-nc-nd https://www.flickr.com/photos/joinash/
Moving away from silos
Acoustic Coupler
Source: “Games Aktuell Blog”, http://www.gamesaktuell.de/Community/MySite/GenX3601966-2605282/Blogs/Cyberpunks-beim-Mauerfall-694794/
Web of Documents (WWW)�Linked Documents
Web of Data / Knowledge
Linked Data
Semantic Web
Tim Berners-Lee
(The inventor of the Web)
(and the Semantic Web)
How does all this work?
Structured data not documents
Graph (networked) data!
W3C Web standards stack
URIs, HTTP, RDF, RDFa, RDFS, OWL, SKOS, SPARQL, etc.
3 principles for linked data/semantic web
Use HTTP IRIs for Things
Internationalised Resource Identifier (IRI) is a string of characters used to identify a name of a resource
http://rijksmuseum.nl/data/painting001
or (rijks:painting1)
Resource Description Framework (RDF)
Semantic Web standard for writing down data, information, knowledge
(Subject, Relation, Object)
<Painting001, has_location, Amsterdam>
Painting001
Amsterdam
has_location
Triples!!
Triples form
rijks:Painting001
geo:Haarlem
dcterms:spatial
dcterms:creator
rijks:Frans_Hals
147590
geo:population
52.38084, 4.63683
geo:partOf
geo:Noord-Holland
geo:Netherlands
geo:coordinates
geo:partOf
ex:Painting002
dcterms:creator
Knowledge Graphs
Relational Databases vs Knowledge Graphs
Relational Database
Knowledge Graph
Knowledge graphs and AI
But, is this all just theory?
KNOWLEDGE GRAPHS EVERYWHERE
http://lod-cloud.net/
(Frank van Harmelen’s Good News slide)
KNOWLEDGE GRAPHS EVERYWHERE
Knowledge graphs
Knowledge graphs are a useful way of representing data, information and knowledge …
Knowledge graphs for Social Good
Why now?
Where AI used to be hidden, with new success and increased societal impact, questions about AI for Good start to emerge..
Example: AI4good at UN
Based on available data a ML model was trained to predict the cost of various conflicts
Identify SDG-related events in social media
Recognizes phrases like “next Saturday …”
Links to SDG
Example: Landportal
Example: Agrovoc/Agris Published by FAO
Example: connecting Aid information in Knowledge Graphs
Msc. Thesis by Kasper Brandt
Network of aid information
IATI:Higher Education
IATI:UN Habitat
Activity: Multi donor fund to support civil society in democracy related issues
Aid received (USD)
Geonames:Gambia
Worldbank:Gambia
N 13° 30' 0'' W 15° 30' 0'
is a
IATI:Sector
IATI:Organization
Geonames:Country
Worldbank:country
is a
is a
is a
End-user application
Kasper Brandt
Can the Knowledge Graphs (be made to) mean something for knowledge sharing even under very constraining conditions?
No internet, no computer, no electricity
Multitude of languages, levels of literacy
Three challenges for Bringing Knowledge Graphs to the Rural Poor
Make KGs usable in low-resource, low-connectivity contexts
Make KGs accessible for users with various (cultural) backgrounds and levels of literacy;
Develop knowledge sharing cases and applications relevant for the rural poor
Infrastructure
Interface
Relevancy
Case: Market information in rural Mali
Knowledge graph includes voice labels
rm:offering0001
rm:shea_butter
rm:product_name
rm:1000
rm:quantity
rdfs:label
rdfs:label
“Amande de Karité”@fr
“Shea Nuts”@en
speakle:voicelabel_ba
rm:audio_shea_nl.wav
rm:audio_shea_ba.wav
speakle:voicelabel_nl
rdfs:label
“1000”
speakle:voicelabel_ba
rm:audio_1000_nl.wav
rm:audio_1000_ba.wav
speakle:voicelabel_nl
rm:Mazankuy_Diarra
rm:kilo
rdfs:label
“kilo”@en
speakle:voicelabel_ba
rm:audio_kilo_nl.wav
rm:audio_kilo_ba.wav
speakle:voicelabel_nl
rm:unit_measure
rm:has_contact
“Slot and Filler” Text-to-Speech
Z_Diarra_ba.wav
offered by.wav
English:
Bambara:
15
liters
of
offered by
Zakari Diarra
15_ba.wav
L_ba.wav
Of_ba.wav
Honey_ba.wav
honey
Talking to Knowledge Graphs
Web applications
<VoiceXML> to SPARQL
Voice browser
RadioMarché
Linked market data
‘Allo Knowledge Graph?
DBpedia
GeoNames
Agrovoc
Challenge: Low-resource Knowledge sharing
High-end hardware, high connectivity
DOWNSCALING
Mid-range hardware, high connectivity
Affordable hardware, various connectivity situations
Kasadaka: Low-resource knowledge sharing platform
Kasadaka (“talking box”)
Andre Baart
Case: DigiVet
Gossa Lo, Myrthe van der Wekken, Romy Blankendaal
Veterinary service in North Ghana
Animal health ontology developed by eliciting information from domain experts using CommonKADS method
Users can interact with system through voice, simple visual interaction
Challenge: Knowledge exchange in low-connectivity setting
GSM network
Semantic Web in an SMS
Enable Semantic Web data exchange over GSM networks.
Converters to translate SPARQL HTTP request to SMS message (140 or 160 chars) and vice versa
Onno Valkering
Evaluation: 4 scenarios in 2 cases
Digivet and RadioMarche applications
Four scenarios / SPARQL queries in total
Results
Lessons learned
Knowledge Graphs allow for knowledge sharing in ICT4D cases
But we need to develop solutions that mean something on the ground
User-centric cases
Technological innovations
Voice-access to knowledge
Semantic Web without the Web
Low-resource computing
Infrastructure
Interface
Relevancy
Polyvocality: �A fourth challenge
The problem
Knowledge graphs are either made from existing sources or by humans…
..therefore they are sure to contain
Biased
Univocal
Single-view
Information, often ased on the majority-view.
This poses the danger of perpetuating biased view on data.
(gender, geographic location, colonial sources,…)
Bias Perspective
Metadata enrichment and bias detection of colonial architecture�Roz Sabir �
2053 missionary objects from Africa
Loot
Gift
Idol
Fetish
Ancestor
PRESSING MATTER
OWNERSHIP, VALUE AND THE QUESTION OF COLONIAL HERITAGE IN MUSEUMS
Towards polyvocal Knowledge Graphs
Identifying and acquiring polyvocality knowledge
Representation of polyvocality: datamodels and formalisms
- Represent disagreement on categorisation, provenance, etc.
Presentation of polyvocal knowledge �- present it to variety of users, including
researchers
heritage professionals
general public
‘source communities’�
Early work: Representing Traditional Knowledge
�1 How is the Sun dance ritual in the Hopi tribe different from the Sun dance ritual by the Cree people? �2 What traditional knowledge is practised in Curacao? |
e.g. 4 Who created the video “How to make�homemade briquettes”?�e.g. 5 What else does Anna know about? �
Lois Hutubessy �
Take home
Knowledge Graphs are the formalism to represent and share Data, Information and Knowledge on the Web
Knowledge Graphs are a part of the AI for Good agenda.
But to make them universally accessible we need to address
and
Thank you
victordeboer.com
v.de.boer@vu.nl