Linked Data Profiling Andrejs Abele National University of Ireland, Galway Supervisor: Paul...

Preview:

Citation preview

Linked Data Profiling

Andrejs Abele

National University of Ireland, Galway

Supervisor: Paul Buitelaar

Overview

Terminology Motivation My approach Evaluation Conclusion Future work

Linked Data is about using the Web to connect related data that was not previously linked.

Resource Description Framework is represented by sets of subject-predicate-object triples, where the elements may be URIs, literals

https://www.insight-centre.org/users/andrejs-ābele foaf:name “Andrejs Ābele”

Linked Open Data Cloud is a collection of Linked Data resources that are open and freely available

Terminology

Linked Open Data Cloud Diagram

Publications

Life Sciences

Cross-Domain

Social Networking

Geographic

Government

Media

User-Generated Content

Linguistics

Motivation

Linked Data is hard to understand for humans Only a small number of datasets provide a

human readable overview or comprehensive metadata

When adding a new dataset to the LOD cloud, connections have to be identified to as many other relevant LOD datasets as possible

LOD Cloud Diagram relays on human classification

Existing solutions for LD profiling

[1] http://demo.seco.tkk.fi/aether/#/ [2] https://www.hpi.uni-potsdam.de/naumann/sites/prolod++/#[3] http://lodlaundromat.org/

[4] http://stats.lod2.eu/ [5] http://demo.seco.tkk.fi/aether/#/[6] http://rdfstats.sourceforge.net/

Loupe1

ProLOD++2

LOD Laundromat3

LODStat4

Aether5

RDF-stats6

Domain identification method using DBpedia

Topic Extraction

Domain Identification

Domain

• Input : Bio2RDF-sgd

• Description: The Saccharomyces Genome Database (SGD) collects and organizes information about the molecular biology and genetics of the yeast Saccharomyces cerevisiae

1. Most frequent terms (sgd_vocabulary, query, proper, phenotype, experiment)

2. Literal containing one of the terms ("protein [sgd_vocabulary:protein]@en")

3. Identify DBpedia concept (http://dbpedia.org/resource/Protein)

4. Identify Category (http://dbpedia.org/resource/Category:Molecular_biology)

5. Identify domain under which category fits best (Biology =>Life Sciences)

Example

DatasetsLOD cloud datasets (annotated in LOD Cloud Diagram)405 datasets, 9 domains • Media (13)• Linguistics(34)• Publications (111)• Social Networking (41)• Geography (29)• Government (65)• Cross Domain (25)• User Generated (52)• Life Sciences (35)

1. Extract URIs of properties and classes from datasets2. Use classes and properties as features3. Classify using Support Vector Machine classifier4. Use Precision and Recall as metrics

Extended baselineEnrich the data with human annotated tags from Linked Open Vocabularies1

1. http://lov.okfn.org/dataset/lov/

Baseline approach

Precision and Recall for different domains using SVM

Media

Linguist

ics

Publicatio

ns

Social n

etwork

ing

Geogra

phy

Gove

rnm

ent

Cross

dom

ain

User g

enerate

d

Life s

cience

s0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

PrecisionRecall

Correctly Classified Instances

Classes Properties Classes + Properties

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

From DatasetDataset+LOVLOV

Conclusion

• Does not require training

• Works with new and customized vocabularies

• Works only if datasets contain literals

• Can not identify User-Generated Content and Cross-Domain

• Using just classes and properties is hard to improve results above 75%

Future Work

• Evaluate alternative classification algorithms

• Use Literals and URIs for classification

• Classify datasets in more specific subdomains

Thank you!

Recommended