Embed Size (px)
DESCRIPTION
Citation preview
Non-Temporal Orderings for Extensional ConceptDrift
Albert Merono-Penuela1,2 Stefan Schlobach1
1Department of Computer Science, VU University Amsterdam, NL
2Data Archiving and Networked Services, KNAW, NL
October 22th, 2013First International Workshop on Semantic Statistics
ISWC 2013
Merono-Penuela, A. et al. Extensional Concept Drift
MotivationStability of Meaning of Concepts
As the world changes continuously, concepts also change theirmeaning over time. We call this concept drift
We call this concept drift
Smooth transitions
Radical transitions (concept shift)
Merono-Penuela, A. et al. Extensional Concept Drift
MotivationTime-Based Stability of Meaning
Concept drift takes place over time. Time is the intuitive orderingof the data.
Dutch historical census data:
1795 1830 1840 1849 1859 1869 1879 1889 1899 1909 1919 19201930 1947 1956 1960 1971
Merono-Penuela, A. et al. Extensional Concept Drift
MotivationTime-Based Stability of Meaning
Concept drift takes place over time. Time is the intuitive orderingof the data.
Dutch historical census data:
1795 1830 1840 1849 1859 1869 1879 1889 1899 1909 1919 19201930 1947 1956 1960 1971
But no time series in the challenge data!
Australia: 2011France: 2010
Merono-Penuela, A. et al. Extensional Concept Drift
MotivationNo-Time Based Stability of Meaning
1 Resignation
2 Discard time for this experiment
No time, no party?
Merono-Penuela, A. et al. Extensional Concept Drift
MotivationNo-Time Based Stability of Meaning
Consider time as just one of all possible dimensions that can beused as data orderings.
Meaningful orderings (i.e. intrinsically dynamic):
GDP per capita
Stock markets
Population density
(Il)literacy
Merono-Penuela, A. et al. Extensional Concept Drift
Case-StudyExperiment Setup
Can we detect drastic extensional drifts over orderings other thantime?
E.g. Is there drift in the concept of youth unemployment1 inAustralia/France as a selected ordering grows?
Australia: GDP per capita per stateFrance: Population density per departement
115-24 years old considered standardMerono-Penuela, A. et al. Extensional Concept Drift
Case-StudyQuerying
SPARQL template for unfolding RDF Data Cubes
Merono-Penuela, A. et al. Extensional Concept Drift
Case-StudyQuerying
Data sources: local endpoint, DBPedia (for the orderings)
Merono-Penuela, A. et al. Extensional Concept Drift
Case-StudyQuerying
Age range Gender Location Occupation Population GDP15-19 years Male New South Wales Unemployed, Total 17456 57828
Age range Gender Location Occupation Population Density15 to 19 years Women Tarn Unemployed 1.345e03 64.17
Merono-Penuela, A. et al. Extensional Concept Drift
Case-StudyResults
1 2 3 4 5 6 7 8
0.0
0.2
0.4
0.6
0.8
1.0
Index
p.gd
p
Drift of youth unemployment inAustralian states using GDP percapita as ordering
0 20 40 60 80
0.0
0.2
0.4
0.6
0.8
1.0
Index
p.de
nsity
.yu
Drift of youth unemployment inFrench departements usingpopulation density as ordering
Merono-Penuela, A. et al. Extensional Concept Drift
Case-StudyResults
2 4 6 8
−4
−3
−2
−1
01
Index
p.gd
p.di
st
Drift of youth unemployment inAustralian states using GDP percapita as ordering (smoothfunction)
0 20 40 60 80
0.95
0.96
0.97
0.98
0.99
1.00
Index
p.de
nsity
.dis
t
Drift of youth unemployment inFrench departements usingpopulation density as ordering(smooth function)
Merono-Penuela, A. et al. Extensional Concept Drift
Conclusions
Concept drift: concepts change over time
No time, but party
Meaningful orderings as valuable as time to study conceptdrift (smooth transition assumption)
Takes full advantage of Semantic Web data
Merono-Penuela, A. et al. Extensional Concept Drift
Thank youQuestions, suggestions?
@albertmeronyohttp://www.cedar-project.nl
http://www.data2semantics.org
Merono-Penuela, A. et al. Extensional Concept Drift
