Implementing Linked Data in Low-Resource Conditions

Implementing Linked Data in Low
Resource Conditions
Caterina Caracciolo, Johannes Keizer
{caterina.caracciolo},{johannes.keizer}@fao.org
Food and Agriculture Organization of the UN
09 September 2015

Goals for Today
• Give you a high level view of what is needed
to do Linked Data
• Identify possible bottlenecks due to working
with little resources
• Based on our experience, give you some
suggestions to overcome those bottlenecks

Our background assumptions
Some restrictions are needed…
• Target audience: small-medium size institutions
– This talk is not meant to be a how-to guide for specific
technical problems, but rather a support grid to plan your
entering the linked open data world
• Target data
– We mainly think of textual data, e.g., list of publications
produced by the institution, catalogues of specimens in
the local museum, factsheets on plants, events organized,
..

Topics for today
• What is a “low-resource” condition
• Open Data and Linked Open Data
• An overview of Linked Data lifecycle
– Bottlenecks in terms of resources
– Our suggestions to overcome them
• The example of Agris

1. IT competencies
• Few IT people, over-busy
• Technology fast moving, nothing taught in
school
• Need personal update
– But working environment may not encourage this
– Or there may be language barriers

2. Other IT/IM/cultural issues
• Competency on legal issues – licenses, litigations?
• “It is my data”, even in the same organization
• Different “cultures” in the same workplace
– Domain specialists “know” the domain and the
data – e.g., the reports they produced - do not
want to spend time with “techy stuff”
– IT/IM people may prefer to spend time to make
better system once, instead of repeating ad-hoc
conversions - would like to standardize more
All may require some investments in time

3. Software
• Outdated operating systems and software
– Because of cost of licenses, or cultural issues

4. Hardware
CPU, memory and technology constraints...

5. Electricity
may be unreliable

5. Electricity
..occasionally available…

5. Electricity
…expensive…

6. Internet connection
may be slow…

6. Internet connection
..dependent on the weather…

The trend
Great attention to data
• Interoperability of data – data that can be
reused = processed in different applications
• Standard and open formats are seen as crucial
to interoperability
• Data made available over the web, for
maximum reuse

Open data in a nutshell
• Like other “open” movements: open and free
• See http://opendefinition.org/
• Especially for government-generated data
• E.g., census, public investments, housing, environment, ..
• A variety of formats used to expose the data
• XLS, CSV, XLM, JSON, PPT, SDMX, ..
• Preference for non-proprietary formats
– Most of the data around is “open”, more or less…
• But, check out if your country has produced a national
policy on data!

Who does Open Data?
• National and regional initiatives (not exhaustive)
– opendataforafrica.org
– data.gov.uk
– usopendata.org
– opendatalatinamerica.org
– open-data.europa.eu
– data.gov.au
– data.gov.in
• Global and sectorial initiatives – e.g., GODAN

Why do people go for Open Data
• Increase transparency of governments and
institutions
• Create new business opportunities
• It is the way to go now

Implementing Linked Data in Low-Resource Conditions

Linked Open Data in a nutshell
• Like other “open” movement: open and free
– You can have Linked Data that with no open license –
but today we think of Linked Open Data (LOD)
• Any type of data, any domain
• The format of choice: RDF
– Various serialization possible – XML, Turtle, N-
Triples, N-Quads, JSON-LD, Notation 3, TriX
• Not just getting datasets out, but linked pieces
of data

Why should I go for Linked Data?
• To be able to reuse data published by others
• To promote business – made by others or
yourself
• Not to be isolated, left behind in the
information world
• Yes but… is the game worth the candle?

Agris - a LOD-based application

Then, Open Data or Linked Data?
• Can be seen as two steps along the same line
• You should decide based on your situation and
goals
– Open data requires less effort. Good if data will be
primarily used by others or have no direct interest
in linking to other datasets
– Linked Open Data may be more complex because
of the linking step. Good if you want to exploit the
data yourself, e.g. to enhance your library/doc rep
catalogue with data produced by others

A typical Linked Data flow
SPARQL endpoint
HTML/RDF
Content negotiation
RDF store
RDF dump
LOD based
applications
Data consumptionLOD exposureLOD storage
“Original “
dataset
Maintenance in
RDF
Maintenance in
original format
Conversion
SPARQL endpoint
“Before” the LOD

RDF
• RDF is simply triples
– Subject – predicate - object
titleID
dct:title
• Triples may be serialized in various formats
– RDF/XML, Turtle, N-triples, N-Quads, JSON-LD, TriX

The role of predicates
• … the dct:title in previous slide, to indicate the
“title” of a book
• Important to expose the data without
ambiguities
• Recommendation is to use standards, or de
facto standard, to facilitate reuse of data
• Search for the vocabulary appropriate to your
data, e.g. with http://lov.okfn.org/dataset/lov/index.html
– Look also at W3C Best Practices for Publishing Linked data
http://www.w3.org/TR/ld-bp/

Conversion from existing formats

Converting data to RDF
• Many converter to RDF
– A list in http://www.w3.org/wiki/ConverterToRdf
• Conversion could be done as a one-time
migration effort, or could be scheduled
regularly
– When done regularly, for exposing your data, your
established data maintenance is not affected

An simple example of conversion

My dummy table
ID book Author Title Subject
1 John Dee Perfect Art of Navigation Navigation,
geography
2 Jethro Tull The new horse-houghing
husbandry
Horse
husbandry

1. Get some RDF
“The perfect Art of Navigation”
John Dee
1
Subject
Title
Author
Navigation

2. Get some linked RDF
“John Dee”
(Agrovoc URI)
<URI>
dct:subject
dct:title
dct:creator
http://aims.fao.org/aos/agrovoc/c_15908

3. Get some more links
http://dbpedia.org/page/John_Dee
(Agrovoc URI)
<URI>
dct:subject
dct:title
dct:creator
http://aims.fao.org/aos/agrovoc/c_15908

Data maintenance
• If data is regularly converted to RDF, the “old”
maintenance flow is kept
– But with the extra step of linking
• If data is once for all migrated RDF, may have
the problem of maintenance – you may need
a GUI

What can be linked?
1. Vocabularies used to describe and annotate
the data - or ontologies
– i.e., the properties of the triples - your “Title”
and somebody else’s “Titulo”
2. The entities linked, the “objects”
– i.e., the object of the triple – a specific author in
your dataset to the same author in somebody
else’s dataset, or in Wikipedia
• Often, they are also called vocabularies,
which may create confusion

1. Linking vocabularies
• It is a research area
– Ontology Alignment Evaluation Initiative (OAEI)
– Note that “ontology” is often used as a generic
term, also to mean rather simple vocabularies to
describe data – ontology may sometimes also
include “individuals”, e.g., country names, ..
• Best solution is to go for standard vocabularies
from the start!
– When you design the conversion of your data

2. Linking “individuals”
• Relatively simple problem, but few out-of-the-
box tools
– Usually the problem is data “cleanliness” – e.g.,
different name spelling, abbreviations, …
• Best solution is to identify the top dataset(s)
to link and start linking to it/them
– Either manually or semi-automatically (Automatic
selection of candidate links, then manual check)
– Data validation usually outside the rest of the data
lifecycle

Hint: Drupal for your catalogue

Drupal = a content management
system
• Allows you to:
1. import data from csv, xml, RSS feed
2. create RDF
3. maintain the data from GUI
4. expose RDF
• Good for your catalogues of documents,
people, ..
• Need to know Drupal, but no programming
skills required

Similar tools
• AgriDrupal
– Drupal customized for small institutions
– Includes tools for automatic tagging with
AGROVOC, which is a linked resource
• ScratchPad
– Customized for biodiversity data

If you want to have your thesuarus
linked…
• This is our experience - AGROVOC
• Thesauri are used for document indexing
(dct:subject “navigation”)
• Steps:
– Convert the thesaurus into SKOS concept scheme
– Use VocBench for data maintenance, including
links
– Use SKOSMOS for data visualization and search

Triple stores
• Very many around, also very many benchmark
to compare performances and functionalities
– Cf. http://www.w3.org/wiki/RdfStoreBenchmarking
• Some tech know-how needed to choose the
best solution and keep it up and running

Various options
1. Provide a dump for download
2. Expose de-refenceable URIs
3. Expose sparql endpoint
4. Expose webserivces

RDF dump for download
• Pros
– Simply a file to download
– For data consumers, access to data is under control
-> efficient, fast
• Cons
– The issue may be to keep the dump in synch
– Need to decide policy on versioning
– Need to decide what to include in the dump (only
the data? Also the links? ..)

De-referenceable URIs
• Pros:
– Data exposed is always up-to-date
– Serving content for URIs
– Simple back-ends are available to visualize also the
html - e.g. Pubby, Loddy
• Cons:
– Need to set up content negotiation mechanism.
Not a big issue, but server must be up 24/7..
– Data is accessible but not searchable by humans

SPARQL endpoint
• Pros:
– Not much work involved, typically endpoint is
provided by triple store
• Cons:
– Require 24/7 server availability
– No limitations on queries -> may be heavy on
server side
• Other solutions under study, e.g.
http://linkeddatafragments.org

Web Services
• Pros:
– Known technology, good performances
– More control on data access, less strain on server
– May be built on top RDF store
• Cons:
– Need to be implemented

Multilingual vocabularies can help

In practice…
An institution with limited
resources wants to move to Linked
Data. What to do?

You have at least two options
1. Consider your specific bottlenecks and go
ahead on your own
2. Organize a collaboration
– Effort on creating partnership, networks

AGRIS
An example of collaborative
approach to LOD

The AGRIS network
Data coordination
Partner
Partner
Partner
Partner
Partner
Partner
Can be much smaller o bigger!
Partner
Partner

……a bibliographical record original

…the same record in a mashup page
http://agris.fao.org/agris-search/search.do?recordID=QM2007000047

26%
22%
14%
11%
9%
4%
3%
2%
2%
2%
1% 1%
3%
Metadata tools used by AGRIS Providers
WebAgris AMM OJS Mendeley WebAGRIS
PubMed InMagic DOAJ GFIS system Dspace
AgriDrupal RISC Others

…using the journal name or the ISSN

…using aligments between thesauri

http://agris.fao.org/agris-search/search.do?recordID=PL2009000495

http://agris.fao.org/agris-search/search.do?recordID=PH2011000084

1. Understand your own
constraints

2. Keep an eye on tech
improvements

In brief…
• Start small: one dataset only (or few)
• Start relevant: choose a key dataset, either
because central to your application, or
because widely used (visibility)
• Start from somewhere: try to reuse
experience as much as possible
• Go in steps: open first, then link
• Look for collaborations

Find your own union
• Organize a consortium and maximize your
resources
• Look for experience and support from other
organizations

Thank you!
caterina.caracciolo@fao.org
johannes.keizer@fao.org
http://aims.fao.org

Implementing Linked Data in Low-Resource Conditions

More Related Content

What's hot (20)

Similar to Implementing Linked Data in Low-Resource Conditions (20)

More from AIMS (Agricultural Information Management Standards) (20)

Recently uploaded (20)

Implementing Linked Data in Low-Resource Conditions