Leveraging the Web of Data: Managing, Analysing and Making Use of Linked Open Data

Institute for Web Science & Technologies – WeST
Leveraging the Web of Data:
Managing, Analysing and Making
Use of Linked Open Data
Thomas Gottron
July 18th, 2014
IRSS

Thomas Gottron IRSS, Athens, 18.7.2014, 2Leveraging the Web of Data
Linked Open Data – Vision of a Web of Data
§  „Classic“ Web
w  Linked documents
§  Web of Data
w  Linked data entities

§  „Classic“ Web
Linked Open Data – Vision of a Web of Data
§  Web of Data
ID
ID

LOD – Base technologies
§  IDs: Dereferencable HTTP URIs
§  Data Format: RDF
§  No schema
§  Links to other data sources
foaf:Document
„Extracting schema ...“
fb:Computer_Scientist
dc:creator
http://dblp.l3s.de/.../NesterovAM98
http://dblp.l3s.de/.../Serge_Abiteboul
rdf:type
„Serge Abiteboul“
dc:title
rdf:type
foaf:name
http://www.bibsonomy.org/.../Serge+Abiteboul
rdfs:seeAlso
1 Statement = 1 Tripel
Subject Predicate Object
rdf:type = http://www.w3.org/1999/02/22-rdf-syntax-ns#type
foaf:Document = http://xmlns.com/foaf/0.1/Document
swrc:InProceedingsrdf:type

LOD Cloud
… the Web of Linked Data consisting of
more than 30 Billion RDF triples from
hundreds of data sources …
Gerhard Weikum
SIGMOD Blog, 6.3.2013
http://wp.sigmod.org/
Where’s the Data in
the Big Data Wave?

Some „Bubbles“ of the LOD Cloud

Making Use of the Linked Data Cloud ...
Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/
LOD: a rich, huge, diverse, public and distributed knowledge base on the Web.
Pros Cons
rich
knowledge
base
diversepublic
huge
on the Web
diversedistributed
Find technical solutions to
overcome challenges

Managing
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Search
data
structure
Efficientstorage
andretrieval
MakingUseAnalysing
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1
Week of Data Snapshot
RDF Type TS PS IPS ECS SchemEX
E1
rdf:type dc:creator
E2
Bad News ...dc:title
foaf:Document
swrc:InProceedings
rdf:type

Managing
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Search
data
structure
Efficientstorage
andretrieval
E1
rdf:type dc:creator
E2
foaf:Document
swrc:InProceedings
rdf:type
Analysing
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1
MakingUse

Data Format
§  Linked Data as N-Quads:
triple – what is the information?
context URI – where does it come from?
s op
c
( )s op c

Index Models

(Abstract) Index Models
w  D : Data elements to be retrieved (payload)
w  K : Key elements to access the data (index elements)
w  σ : Selection function: How to get data for a key
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
DK σ
Searchdata
structure
Efficientstorage
andretrieval
℘( )
Data items / PayloadKeys

Choices for Key Elements
Subject
s op c
Search
data
structure
s1 s2 sn
Context
Search
data
structure
c1 c2 cm
s op c
Literals
Search
data
structure
x y z
s litp c
Types
Search
data
structure
t1 t2 tk
s t
rdf:
type c
...
Alternative:
predicates or
objects

Choices for the Payload
Full Caching
local
Web
s op c
Triples
local
Web
s op
Entities
local
Web
s
Data Sources
local
Web
c
...

, ...
Concrete Example: Subject Based Index Model
west:Gottron
west:Staab
west:Schegi
...
tud:CGottron
(west:Gottron, rdf:type, foaf:Person)
(west:Gottron, foaf:knows, west:Staab)
...
(west:Staab, swrc:institution, west:WeST)
(west:Staab, foaf:name, „Steffen Staab“)
...
(west:Schegi, rdf:type, foaf:Person)
(west:Schegi, foaf:name, „Stefan Scheglmann“)
(tud:CGottron, swrc:institution, tud:KOM)
(tud:CGottron, foaf:knows, west:Gottron)
...
s o1p1s s o2p2

Implemented Index Models
§  Triple based
§  Meta data
https://github.com/gottron/lod-index-models
s ops
s opp
s opo
s opterm
s opc
s opPLD

Schema-level Indices

Schema Information on the LOD Cloud
(No)
Schema?
Guidelines / best practices
Automatic tools Social effects
Emerging
Schema!
Induce from data
observations

Examples for Schema Information
Property Set Type Set
{ }
...
x, ...p1 p3p2 { }
...
t1 t2 y, ...
rdf:type
y
rdf:type
t1
t2
x
p1
p2
p3

Implemented Index Models
§  Triple based
§  Meta data
§  Schema-level
https://github.com/gottron/lod-index-models
s ops
s opp
s opo
s opterm
s opc
s opPLD
type s
SchemEX s
t
t st t
p
p sp p
p-1
p-1 op-1p-1
t
p sp t

SchemEX

Schema-based Access to the LOD cloud
?
foaf:Document
dc:creator
x
swrc:InProceedings
SELECT ?x
WHERE {
?x rdf:type foaf:Document .
?x rdf:type swrc:InProceedings .
?x dc:creator ?y .
?y rdf:type fb:Computer_Scientist
}

Schema-
level Index
Where?
•  ACM
•  DBLP
SELECT ?x
WHERE {
?x dc:creator ?y .
}
Which schema
information?

Typecluster
§  Entities with the same Set of types
t1 t2
C1 C2 Cm
tn...
...
TCj

Typecluster: Example
foaf:Document swrc:InProceedings
DBLP ACM
tc2309

Property Sets
§  Entities with the same Set of properties
p1 p2
C1 C2 Cm
pn...
...
PSi

Bi-Simulation: Example
dc:creator
BBC DBLP
ps2608

SchemEX: Combination TC and PS
§  Partition of TC based on PS with restrictions on the
destination TC (equivalence relation)
t1 t2 tn...
C1 C2 Cm
...
t45 t2 tn‘...
p1 pn‘‘...
EQC EQC
Cx
TCj TCk
EQCl
PSi
SchemaPayload
p2
PS

SchemEX: Example
DBLP
...
tc2309 tc2101
eqc707
ps2608
foaf:Document swrc:InProceedings fb:Computer_Scientist
dc:creator
SELECT ?x
WHERE {
?x dc:creator ?y .
}

Building Indices

Building Indices: Operators
§  Combination of few simple operations
w  Aggregate, Join, Invert
§  Example: Property Set index
s1 o1p1 c1
s1 o1p2 c1
s2 o2p2 c1
s3 o3p1 c1
s3 o4p2 c1
s4 o1p3 c1
s1 p1 p2
s2 p2
s3 p1 p2
s4 p3
p1 p2 s1 s3
p2 s2
p3 s4
Aggregate Invert

SchemEX: Computation
§  Precise computation: 3 Aggregates + Join + Invert
t1 t2 tn...
C1 C2 Cm
...
t45 t2 tn‘...
p1 pn‘‘...
EQC EQC
Cx
TCj TCk
EQCl
PSi
SchemaPayload
p2
PS
Better Approach?(Faster, more scalable)

Stream-based Computation of SchemEX
§  LOD Crawler: Stream of N-Quads
… Q16, Q15, Q14, Q13, Q12, Q11, Q10, Q9, Q8, Q7, Q6, Q5, Q4, Q3, Q2, Q1
FiFo
4
3
2
1
1
6
2
3
4
5
t3
t2
t2
t1
t1 t2 tn...
C1 C2 Cm
...
t45 t2 tn‘...
p1 pn‘‘...
EQC EQC
Cx
TCj TCk
EQCl
PSi
SchemaPayload
p2
PS

Quality of Approximated Index
§  Stream-based computation vs. precise computation
w  Data set of 11 Mio. tripel

SchemEX: 1st place @ BTC 2011
§  SchemEX
w  Allows complex queries (Star, Chain)
w  Scalable computation
w  High quality
§  Index over BTC 2011 data
w  2.17 billion triple
w  Index: 55 million triple
§  Commodity hardware
w  VM: 1 Core, 4 GB RAM
w  Throughput: 39.500 triple / second
w  Computation of full index: 15h

k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Search
data
structure
Efficientstorage
andretrieval
ManagingMakingUseAnalysing
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1
E1
rdf:type dc:creator
E2
foaf:Document
swrc:InProceedings
rdf:type

Redundancy of
Schema Information

Explicit and Implicit Schema Information on LOD
E1
rdf:type
Explicit
Assigning class types
Implicit
Modelling attributes
dc:creator
E2
foaf:Document
swrc:InProceedings
rdf:type
Class Type
Entity
rdf:type Entity
Property
Entity 2
To which degree?

e.g.
Probabilistic model of schema information
§  Joint Distribution P(T,R) over
w  Type sets: TS
w  Property sets: PS
§  P(T=t,R=r) : probability of resource to
have type set t and property set r
P(T,R)
r1

r2

r3

r4

P(T)

t1

14%
2%
5%
8%
29%

t2

5%
15%
2%
3%
25%

t3

7%
3%
30%
5%
45%

P(R)
26%
20%
37%
17%
foaf:Document
swrc:InProceedings
e.g.
dc:creator
dc:title
Marginal
Distributions
Marginal
Distributions

Estimating probabilities
Data set Triples TS PS
Rest 22.3M 793 7,522
Datahub 910.1M 28,924 14,712
Dbpedia 198.1M 1,026,272 391,170
Freebase 101.2M 69,732 162,023
Timbl 204.8M 4,139 9,619
§  Todo (on large data sets)
w  Determine schema use
w  Aggregate counts
§  „Query“ schema-level index:
§  Data background: segments from BTC‘12
p(t,r) =
d ∈ σ t,r( )
N

Redundancy
To which degree can one type of information (either types or
properties) explain the respective other?
§  Mutual Information
§  Normalized MI:
•  H(T) and H(R) : entropy of the marginal distributions.
I(T, R) = p t,r( )⋅log
p t,r( )
P T = t( )⋅ P R = r( )
"
#
$$
%
&
''
r∈PS
∑
t∈TS
∑
I0 (T, R) =
I T, R( )
min H T( ), H R( )( )

Example: Normalized Mutual Information
P(T,R)
r1

r2

r3

r4

P(T)

t1

14%
2%
5%
8%
29%

t2

5%
15%
2%
3%
25%

t3

7%
3%
30%
5%
45%

P(R)
26%
20%
37%
17%

I0 (T, R) = 0.239
P(T,R)
r1

r2

r3

r4

P(T)

t1

22%
1%
0%
1%
24%

t2

1%
23%
1%
0%
26%

t3

1%
0%
48%
1%
50%

P(R)
24%
24%
49%
2%

I0 (T, R) = 0.766

Normalized Mutual Information on BTC‘12
§  Tendencies:
w  Relatively high redundancy
w  Freebase: (weakly) pre-defined schema
w  Timbl: narrow domain (FOAF profiles)
w  DBpedia: de-centralized schema
Data set I0(T,R)
Rest 0.881
Datahub 0.747
Dbpedia 0.635
Freebase 0.860
Timbl 0.850
1
4
5
2
3

Finding Alternative
Descriptions

Searching for a Suitable Description
SELECT ?x
WHERE {
?x rdf:type foaf:Document
}
SELECT ?x
WHERE {
?x rdf:type foaf:PersonalProﬁleDocument
}
SELECT ?x
WHERE {
?x rdf:type sioc:Post .
}
Declarative
descriptions

Operations on the Declarative Description
Entity
Set
C1
C2
C1
C2
C3
C1
C2
C1
C2
C1sub
C1sup
C2
C1
Add
Delete
Refine
Generalize

Just (Small) Baby Steps ...

Formal Concept Analysis
§  Constructs concepts and their hierarchy from data objects
§  Input (formal context)
w  Set O (the objects)
w  Set A (the attributes)
w  Relation I ⊆ O×A (which object has which attributes)
§  Derivation Operator
w  (Sub)Set of objects
w  (Sub)Set of attributes
Attributes common
to all objects in X
Objects which have
all attributes in Y
X ' := y ∈ A : x, y( ) ∈ I,∀x ∈ X{ }
Y ' := x ∈ O : x, y( ) ∈ I,∀y ∈ Y{ }

Formal Concept Analysis
§  (X,Y) is a formal concept, if X'=Y and Y'=X
w  X is the extent
w  Y is the intent
w  Example: ({1,5,10},{a,c})
§  Partial order for formal concepts:
w  (X1,Y1) ≤ (X2,Y2) if X1 ⊂ X2
w  Equivalent to Y1 ⊃ Y2
w  Example: ({1,10},{a,b,c}) ≤ ({1,5,10},{a,b})
§  Defines a lattice structure on the concepts
Table 1. Example of two formal conte
Object a b c
1 ⇥ ⇥ ⇥
2 ⇥
3 ⇥
4 ⇥ ⇥
5 ⇥ ⇥
6 ⇥ ⇥
7 ⇥
8 ⇥ ⇥
9 ⇥ ⇥
10 ⇥ ⇥ ⇥
⌧
⇠⇡ ⇢;
qqqqqq
MMMMMM
⌧
⇠⇡ ⇢{a}
LLLLL
⌧
⇠⇡ ⇢{b}
rrrrr
LLLLL
⌧
⇠⇡ ⇢{c}
rrrrr
⌧
⇠⇡ ⇢{a, b}
LLLL
⌧
⇠⇡ ⇢{a, c}
⌧
⇠⇡ ⇢{b, c}
rrrr
⌧
⇠⇡ ⇢{a, b, c}

Formal Concept Lattice (extent)
≤
Object a b c
1 ⇥ ⇥ ⇥
2 ⇥
3 ⇥
4 ⇥ ⇥
5 ⇥ ⇥
6 ⇥ ⇥
7 ⇥
8 ⇥ ⇥
9 ⇥ ⇥
10 ⇥ ⇥ ⇥
⌧
⇠⇡ ⇢;
qqqqqq
MMMMMM
⌧
⇠⇡ ⇢{a}
LLLLL
⌧
⇠⇡ ⇢{b}
rrrrr
LLLLL
⌧
⇠⇡ ⇢{c}
rrrrr
⌧
⇠⇡ ⇢{a, b}
LLLL
⌧
⇠⇡ ⇢{a, c}
⌧
⇠⇡ ⇢{b, c}
rrrr
⌧
⇠⇡ ⇢{a, b, c}

Formal Concept Lattice (intent)
Top-Concept:
(O,Ø)
Bottom-Concept:
(Ø,A)
Object a b c
1 ⇥ ⇥ ⇥
2 ⇥
3 ⇥
4 ⇥ ⇥
5 ⇥ ⇥
6 ⇥ ⇥
7 ⇥
8 ⇥ ⇥
9 ⇥ ⇥
10 ⇥ ⇥ ⇥
⌧
⇠⇡ ⇢;
qqqqqq
MMMMMM
⌧
⇠⇡ ⇢{a}
LLLLL
⌧
⇠⇡ ⇢{b}
rrrrr
LLLLL
⌧
⇠⇡ ⇢{c}
rrrrr
⌧
⇠⇡ ⇢{a, b}
LLLL
⌧
⇠⇡ ⇢{a, c}
⌧
⇠⇡ ⇢{b, c}
rrrr
⌧
⇠⇡ ⇢{a, b, c}
9 ⇥ ⇥
10 ⇥ ⇥ ⇥
9 ⇥ ⇥
10 ⇥ ⇥
⌧
⇠⇡ ⇢;
qqqqqq
MMMMMM
⌧
⇠⇡ ⇢;
ppppppp
⌧
⇠⇡ ⇢{a}
LLLLL
⌧
⇠⇡ ⇢{b}
rrrrr
LLLLL
⌧
⇠⇡ ⇢{c}
rrrrr
⌧
⇠⇡ ⇢{x}
⌧
⇠⇡ ⇢{y}
qqqqq MMMM
⌧
⇠⇡ ⇢{a, b}
LLLL
⌧
⇠⇡ ⇢{a, c}
⌧
⇠⇡ ⇢{b, c}
rrrr
⌧
⇠⇡ ⇢{x, y}
MMMM
qqq
⌧
⇠⇡ ⇢{a, b, c}
⌧
⇠⇡ ⇢{x, y, z}
Formal concept lattice structures based on the relations in Ta
d by their intent—which provides a better overview.

Navigating the Lattice
Remove constraints
Extend object set
Add constraints
Reduce object set
Nice formalization, but ...
9 ⇥ ⇥
10 ⇥ ⇥ ⇥
9 ⇥ ⇥
10 ⇥ ⇥
⌧
⇠⇡ ⇢;
qqqqqq
MMMMMM
⌧
⇠⇡ ⇢;
ppppppp
⌧
⇠⇡ ⇢{a}
LLLLL
⌧
⇠⇡ ⇢{b}
rrrrr
LLLLL
⌧
⇠⇡ ⇢{c}
rrrrr
⌧
⇠⇡ ⇢{x}
⌧
⇠⇡ ⇢{y}
qqqqq MMMM
⌧
⇠⇡ ⇢{a, b}
LLLL
⌧
⇠⇡ ⇢{a, c}
⌧
⇠⇡ ⇢{b, c}
rrrr
⌧
⇠⇡ ⇢{x, y}
MMMM
qqq
⌧
⇠⇡ ⇢{a, b, c}
⌧
⇠⇡ ⇢{x, y, z}
Formal concept lattice structures based on the relations in Ta
d by their intent—which provides a better overview.

… still Baby Steps

Parallel Lattices
§  Availability of several attribute sets
w  Facet dimensions
w  „Natural“ subdivision
w  Different descriptions of the same data

Parallel Lattices

General Idea for Mapping
Entity
Set
C1
C2
C3
C4
C5
Approx.
Entity
Set
derive
derive
approximate
description alternative
description

Implementing Mappings
§  Minimal Extension
w  Top-Down
Maximal Reduction
Bottom-Up
{b,c}' = {1,4,6,9,10}
Alternative description for {b,c}
1
1,9
10
2,8 3,5,7
Precision? Recall?

Observations
§  On LOD: Mapping type sets onto property sets
w  Evaluation on 20 data sets (subset of BTC‘12)
§  Quality of approximations
w  max-red:
•  High recall: mainly > 0.8
•  Better for smaller concepts
w  min-ext:
•  Good precision: mainly > 0.5
•  Better for larger concepts
rss:Item
sioc:MicroblogPost
foaf:maker
sioc:has_discussion
dcterms:date

Evolution of Linked Data

Evolution of LOD
2007
2008
2009
2010
2011

Evolution of LOD
Time
Volume
Triples provided by data sources
Insertion, deletion,
modification

Effects on Indices: Decline in accuracy
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1

Updates of Indices and Caches
Which sources to
prioritise in an
update?

Change Metrics
§  Comparison of two RDF data sets (e.g. from different
points in time)
w  Xi : Set of triple statements
w  Numeric expression for „distance“
§  Example:
X1
X2
Δ 0,∞[ )
ΔJaccard X1, X2( )=1−
X1 ∩ X2
X1 ∪ X2
Suitable to measuredynamics???

Toy example: Changes Analysis of LOD
1st snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata

1st snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
2nd snapshot
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
Paluno

Changes detected between 1st and 2nd snapshot
1.  Deleted: <InstituteWEST hasMember Gerd>
2.  New: <InstitutePaluno hasMember Gerd >
1st snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
2nd snapshot
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
Paluno

1st snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
2nd snapshot
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
Paluno
3rd snapshot
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata

1st snapshot 2nd snapshot 3rd snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
Paluno
Changes detected between 2nd and 3rd snapshot
1.  New: <InstituteWEST hasMember Gerd>
2.  Deleted: <InstitutePaluno hasMember Gerd >

1st snapshot 2nd snapshot 3rd snapshot
GerdInstitute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
ZBW
Institute
WeST
Thomas
Gerd
Ansgar
Renata
Institute
Paluno
Changes detected between 1st and 3rd snapshot
None!?!
Change metrics
capture differences
We want to
measure
dynamics!

Measuring Dynamics: Requirements
§  Dynamics function Θ
w  quantify the evolution of a dataset X over a period of time
Θti
tj
(X) = Θ(Xtj
)−Θ(Xti
) ≥ 0
Θ
Dynamics as
amount of
evolution
Timeti
tj
X

Constructing a Dynamics Function
§  Function Θ difficult to define directly
§  Indirect definition over a change rate function c(Xt)
Θ(Xtj
)−Θ(Xti
) = c Xt( )
ti
tj
∫ dt
Time
Θ
c
ti
tj
X

Change Rate Function
§  Also c(Xt) not explicitely known!
§  But can be approximated!
w  Given snapshots of the data in small time intervals:
w  The change rate can be approximated via change metrics:
Δ Xti
, Xti−1
( )
ti −ti−1
ti−1→ti
$ →$$ c Xti
( )=
d
dt
Θ(Xti
)
Xt1
, Xt2
, Xt3
,!, Xtn

Dynamics Framework
§  Approximating c(Xt) as step function
Timeti
tj
Θ
c
Θt1
tn
(X) = Δ Xti
, Xti−1
( )
i=2
n
∑ Choice of Δ:
Flexible use of
different notions
of change!
X

Introduction of Decay
§  So far:
w  Impact of evolution independent of moment in time
w  Desirable: Focus on certain periods of time
•  e.g. recent past
§  Solution:
w  Decay function f to assign weights to moments in time
Time
c
ti tj
f
f ⋅c

Implementing a Decay Function
§  Exponential decay function:
§  Incoporated in the framework:
§  When using the step function approximation of c(Xt) :
f t( )= e−λt
Θ(Xtj
)−Θ(Xti
) = e
−λ tj −t( )
⋅c Xt( )
ti
tj
∫ dt
Θt1
tn
(X) = e
−λ tn−ti( )
⋅Δ Xti
, Xti−1
( )
i=2
n
∑

Tabelle1
2012-05-06
2012-06-03
2012-07-01
2012-07-29
2012-08-26
2012-09-23
2012-10-21
2012-11-18
2012-12-16
2013-01-13
2013-02-24
2013-03-24
2013-04-22
2013-05-19
2013-06-16
2013-07-14
2013-08-11
2013-09-08
2013-10-06
2013-11-03
0
0,2
0,4
0,6
0,8
1
Change Rate Function of Seleted Data Sources
Tabelle1
2012-05-06
2012-05-27
2012-06-17
2012-07-08
2012-07-29
2012-08-19
2012-09-09
2012-09-30
2012-10-21
2012-11-11
2012-12-02
2012-12-23
2013-01-13
2013-02-19
2013-03-10
2013-03-31
2013-04-22
2013-05-12
2013-06-04
2013-06-23
2013-07-14
2013-08-04
2013-08-25
2013-09-15
2013-10-06
2013-10-27
2013-11-17
0
0,2
0,4
0,6
0,8
1
Θ = 55.71 , Θdecay = 23.42
dbpedia.org
Tabelle1
2012-05-06
2012-06-03
2012-07-01
2012-07-29
2012-08-26
2012-09-23
2012-10-21
2012-11-18
2012-12-16
2013-01-13
2013-02-24
2013-03-24
2013-04-22
2013-05-19
2013-06-16
2013-07-14
2013-08-11
2013-09-08
2013-10-06
2013-11-03
0
0,2
0,4
0,6
0,8
1
Θ = 58.45 , Θdecay = 18.48
identi.ca
Θ = 51.75 , Θdecay = 25.03
linkedct.org
Tabelle1
2012-05-06
2012-06-03
2012-07-01
2012-07-29
2012-08-26
2012-09-23
2012-10-21
2012-11-18
2012-12-16
2013-01-13
2013-02-24
2013-03-24
2013-04-22
2013-05-19
2013-06-16
2013-07-14
2013-08-11
2013-09-08
2013-10-06
2013-11-03
0
0,2
0,4
0,6
0,8
1
Θ = 20.90 , Θdecay = 8.33
dbtune.org

k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Search
data
structure
Efficientstorage
andretrieval
Managing
E1
rdf:type dc:creator
E2
foaf:Document
swrc:InProceedings
rdf:type
Analysing
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1
MakingUse

Schema-level
Search on LOD

Schema-
level Index
Where?
•  ACM
•  DBLP
SELECT ?x
WHERE {
?x dc:creator ?y .
}

LODatio: Schema-level Search of LOD

LODatio: Query transformation

LODatio: Query transformation
foaf:Document
dc:creator
x
swrc:InProceedings
DBLP
...
tc2309 tc2101
eqc707
ps2608
dc:creator
SELECT ?x
WHERE {
?x dc:creator ?y .
}
SELECT ?c
WHERE {
?eqc schemex:hasDataset ?c .
?tc_A schemex:hasSubset ?eqc .
?tc_A schemex:hasClass foaf:Document .
?tc_A schemex:hasClass swrc:InProceedings .
?bs void:subjectsTarget ?eqc .
?bs void:objectsTarget ?tc_B .
?bs void:property dc:creator .
?tc_B schemex:hasClass fb:Computer_Scientist
}

LODatio: Retrieval Results

LODatio: Retrieval Results
C1
EQCl
C1
EQCl
DS23
URI 1
URI 2
URI 3
Entity count
Example entities

LODatio: User Support

LODatio: User Support
§  Currently implemented:
w  Moderate reductions /
extensions
§  Next release:
w  Include alternative
description based on
parallel lattices
further properties
further types
DBLP
...
tc2309 tc2101
eqc707
ps2608
dc:creator
DS23

LODatio: next steps
Keyword search
Better
recommendations
Other payload
entities
Visual exploration
Related
datasources
Coverage

Focused Exploration
(work in progress)

Use Case: Social Media Coverage of Events

LinkedGeoData
OSM
owl:sameAs
???
Other locations?

Extending LinkedGeoData
Seed Exploration Overlay

Task of Focused Exploration (use case: locations)
§  Prioritise/select object URIs for exploration
umbel:Village
s
-1.404
50.897
wgs84:long
wgs84:lat
dbponto:isPartOf
dbponto:wikiPageExternalLink
dbponto:governmentType
dbpprop:settlementType
dbpprop:subdivisionName
o1
dbpprop:postalCode
dcterms:subject
o4
o5
o6
o7
o8
o10
o9
o11
o2
o3
xxx
yyy
wgs84:long
wgs84:lat

Exploration based on Schema Semantics
§  Exploit rdfs:range definitions of predicates
§  Follow edges which lead to locations with higher priority
dbponto:twinCity dbpedia:City
rdfs:range
dbpedia:Place
rdfs:subClassOf

Supervised Machine Learning
§  Use incoming predicates as features
w  Learn predicates typically leading to locations
§  Train a classifier (e.g. Naive Bayes)
o
xxx
yyy
wgs84:long
wgs84:lat
p2
p3
o‘
p4
p6

IR Inspired Approaches
§  Model discriminativeness of predicates
w  Inspired by tf-idf
§  Property relevance frequency (prf):
•  Normalised version (prr)
§  Inverse property frequency
§  Rank by combine measure: prf-ipf
prf = c(p, L)
ipf = log
c(∗,∗)
c(p,∗)
"
#
$
%
&
'

Performance
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
ROC
random
Schema Semantics
NB (all predicates)
NB (present predicates)
prf-ipf
prr-ipf
0.95
0.975
1
0 0.025 0.05

Performance
. Average performance of approaches († indicates significant improvem
nce level ⇢ = 0.01)
Method Recall Precision F1 Accuracy AUC
Schema Scemantics 0.1188 0.8119 0.2073 0.7262 0.5552
NB (all predicates) 0.9906 0.9491 † 0.9694 † 0.9812 0.9970
NB (observed predicates) 0.9943 0.9436 0.9683 0.9804 0.9968
prf-ipf 0.8512 † 0.9754 0.9091 0.9487 0.9958
prr-ipf † 0.9973 0.9240 0.9592 0.9745 0.9769
mance in bold. Furthermore, we marked the results where we had a signific
ent over the second best method at confidence level of ⇢ = 0.01. The agg
basically confirm the observations made above. In general, when conside
es F1, Accuracy and AUC, the Naive Bayes classifier making use of al

E1
rdf:type dc:creator
E2
foaf:Document
swrc:InProceedings
rdf:type
Analysing
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80
MicroAvg.F1
MakingUse
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Search
data
structure
Efficientstorage
andretrieval
Managing

Summary
Pros Cons
rich
knowledge
base
diversepublic
huge
on the Web
diversedistributed
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...
Technical solutions to
some of the problems

Summary
Pros Cons
rich
knowledge
base
diversepublic
huge
on the Web
diversedistributed
k1
k2
k3
...
kn
d1,1 d1,2 d1,3 ...
d2,1 d2,2
d3,1 d3,2 d3,3 ...
dn,1 dn,2 dn,3 ...

Thank you!
Contact:
Thomas Gottron
WeST – Institute for Web Science and Technologies
Universität Koblenz-Landau
gottron@uni-koblenz.de

References
1.  M. Konrath, T. Gottron, and A. Scherp, “Schemex – web-scale indexed schema extraction of linked open
data,” in Semantic Web Challenge, Submission to the Billion Triple Track, 2011.
2.  M. Konrath, T. Gottron, S. Staab, and A. Scherp, “Schemex—efficient construction of a data catalogue
by stream-based indexing of linked data,” Journal of Web Semantics, 2012.
3.  T. Gottron, M. Knauf, S. Scheglmann, and A. Scherp, “Explicit and implicit schema information on the
linked open data cloud: Joined forces or antagonists?,” Tech. Rep. 06/2012, Institut WeST, Universität
Koblenz-Landau, 2012.
4.  T. Gottron and R. Pickhardt, “A detailed analysis of the quality of stream-based schema construction on
linked open data,” in CSWS’12: Proceedings of the Chinese Semantic Web Symposium, 2012.
5.  T. Gottron, A. Scherp, B. Krayer, and A. Peters, “Get the google feeling: Supporting users in finding
relevant sources of linked open data at web-scale,” in Semantic Web Challenge, Submission to the
Billion Triple Track, 2012.
6.  T. Gottron, A. Scherp, B. Krayer, and A. Peters, “LODatio: Using a Schema-Based Index to Support
Users in Finding Relevant Sources of Linked Data,” in K-CAP’13: Proceedings of the Conference on
Knowledge Capture, 2013.
7.  T. Gottron, M. Knauf, S. Scheglmann, and A. Scherp, “A Systematic Investigation of Explicit and Implicit
Schema Information on the Linked Open Data Cloud,” in ESWC’13: Proceedings of the 10th Extended
Semantic Web Conference, 2013.
8.  J. Schaible, T. Gottron, S. Scheglmann, and A. Scherp, “LOVER: Support for Modeling Data Using
Linked Open Vocabularies,” in LWDM’13: 3rd International Workshop on Linked Web Data Management,
2013.
9.  R. Dividino, A. Scherp, G. Gröner, and T. Gottron, “Change-a-LOD: Does the Schema on the Linked
Data Cloud Change or Not?,” in COLD’13: International Workshop on Consuming Linked Data, 2013.

References
10.  T. Gottron, M. Knauf, and A. Scherp, “Analysis of schema structures in the linked open data graph based
on unique subject uris, pay-level domains, and vocabulary usage,” Distributed and Parallel Databases,
pp. 1–39, 2014.
11.  T. Gottron and C. Gottron, “Perplexity of index models over evolving linked data,” in ESWC’14:
Proceedings of the Extended Semantic Web Conference, 2014.
12.  T. Gottron, A. Scherp, and S. Scheglmann, “Providing alternative declarative descriptions for entity sets
using parallel concept lattices,” in ESWC’14: Proceedings of the Extended Semantic Web Conference,
2014.
13.  Carothers, G.: Rdf 1.1 n-quads. W3C Recommendation (Feb 2014), http://www.w3. org/TR/2014/REC-n-
quads-20140225/, (accessed 14 March 2014)
14.  Käfer, T., Abdelrahman, A., Umbrich, J., O’Byrne, P., Hogan, A.: Observing linked data dynamics. In:
The Se- mantic Web: Semantics and Big Data, Lecture Notes in Computer Science, vol. 7882, pp. 213–
227. Springer Berlin Heidelberg (2013)
15.  T. Gottron, “Of Sampling and Smoothing: Approximating Distributions over Linked Open Data,” in
PROFILES’14: Proceedings of the Workshop on Dataset ProfiIling and Federated Search for Linked
Data, 2014.
16.  R. Dividino, T. Gottron, A. Scherp, and G. Gröner, “From Changes to Dynamics: Dynamics Analysis of
Linked Open Data Sources,” in PROFILES’14: Proceedings of the Workshop on Dataset ProfiIling and
Federated Search for Linked Data, 2014.
17.  R. Dividino, A. Kramer, and T. Gottron, “An Investigation of HTTP Header Information for Detecting
Changes of Linked Open Data Sources,” in ESWC’14: Proceedings of the Extended Semantic Web
Conference, 2014.

Sources
•  Photograph of three of Nevins Memorial Library's earliest librarians. Wikimedia Commons collection,
http://commons.wikimedia.org/wiki/File:Nevins_Library_First_Librarians.jpg
•  Wide-angle view of the ALMA correlator, This Wikipedia and Wikimedia Commons image is from the
European Southern Observatory (ESO) and is freely available at http://commons.wikimedia.org/wiki/
File:Wide-angle_view_of_the_ALMA_correlator.jpg under Creative Commons Attribution 3.0
Unportedlicense.
•  Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/, This
work is available under a CC-BY-SA license.

Leveraging the Web of Data: Managing, Analysing and Making Use of Linked Open Data

Recommended

More Related Content

What's hot (14)

Viewers also liked (20)

Similar to Leveraging the Web of Data: Managing, Analysing and Making Use of Linked Open Data (20)

More from Thomas Gottron (9)

Recently uploaded (20)

Leveraging the Web of Data: Managing, Analysing and Making Use of Linked Open Data