LOD云中身份和重定向的演变审视

Shuai

Wangwith

Idries

Nasim,

Joe

Raad,

Peter

Bloem,

Frank

van

HarmelenWAI,

19th

SeptemberExamining

the

Evolution

of

Identity

andRedirection

in

the

LOD

CloudhLps:///shuaiwangvu/idenPty_graph_evoluPonhLps:///shuaiwangvu/redirecPon•2Introduction

and

related

workEvolution

of

identity

graphsConstructing

the

new

identity

graphsCompare

the

identity

graphsAnalysis

of

redirectionConstructing

the

redirect

graphsA

qualitative

analysisA

quantitative

analsysisDiscussion

and

future

work•••OUTLINE•Identity

crisis

[Halpin,

et

al.]17.9%

entities

in

DBpedia

do

not

exist

after

2

years

[De

Melo,

2013]Semantic

web

evolution

[multiple]Semanticallybroken

links

[Regino

et

al]•••INTRODUCTION

AND

RELATED

WORKBlack

=

identity

linksRed

=

redirectionBlue

=

encoding

equivalence3RQ1:

How

has

the

identitygraph

in

the

semantic

web

changed?RQ2:

Can

graphs

of

redirects

provide

an

indication

of

the

evolution

of

the

identitygraphs

in

the

semantic

web?RQ3:

Can

we

approximate

the

implicit

semantics

of

redirection?RQ4:

What

are

the

properties

of

the

redirection

graph?4RESEARCH

QUESTIONSLOD

Laundromat

in

2015

shows

that•91.2%

of

entities

are

in

linksets8.1%

of

entities

are

in

major

hubs

with

more

than10identity

links.Construct

the

new

identity

graph

with•linksets:

DBpedia

DatabusMajor

linked

data

hub:

Yago,

Pleiades,

WordNet,etc.CONSTRUCTING

THE

NEW

IDENTITY

GRAPHS5•HTTP

200:

‘OK’400+

HTTP

error:

Not

Found:

‘NF’a

literal

or

the

request

fails:

‘ER’Timeout:

‘TO’••••All

redirects

of

300+Redirects

Until

Found:

‘RUT’RedirectUntil

Not

Found:

‘RUNF’Redirect

Until

Error:

‘RUE’Redirect

Until

Found:

‘RUF’••••657.9M

entities

are

shared=

32.3%

of

Gand13.4%

of

H.H

consists

of

many

more

entitiesthan

G.The

triple:entity

ratio

has

droppedfrom

3.12

in

G

to

0.94

in

H,

whichindicate

that

redundant

edgesmight

be

fewer

in

H.The

HDT

file

of

I

is

3.3

times

biggerthan

G.7COMPARING

THE

GRAPHS#

Triples#EntitiesSize

(HDT

file)G

(old

graph)558M179M4.5GH

(new

graph)409M443M11GI

(integrated

graph)951M555M15GFor

the

largest

CC

of

I.293700

(28%)

nodes

fromG,

450107

(44%)

from

H,290877(28%)

from

both•37176(46,5%)

CC’s

fromG,

42718(53,4%)

CC’sfrom

HCOMPARING

THE

GRAPHS|Biggest

CC|#CCSize

(HDT

file)G

(old

graph)178K49M4.5GH

(new

graph)219K137M11GI

(integrated

graph)1M164M15G8For

H

and

G:Sampling

100K

uniformlySampling

20K

fromCC(2)CC(3-10)CC(>10)CONSTRUCTING

THE

REDIRECT

GRAPHSDenoted9COMPARING

THE

OLD

AND

NEW

GRAPHS10Valid

=

OK

+

RUF(redirect

untilfound)Invalid

=

the

restANALYSIS

OF

THE

REDIRECTION

GRAPH11G

has

more

validentities

than

H.Only

1-3%

returnsmeaningful

infodirectly.>50%

has

redirectionfor

uniform

samplingANALYSIS

OF

THE

REDIRECTION

GRAPH12#Valid

decreases

as

the

sizeof

CCs

increase,

especially

H.Opposite

trend

for

NF,

TO,RUNF,

RUEDifferent

for

OK

is

too

smallto

draw

a

conclusion.ANALYSIS

OF

THE

REDIRECTION

GRAPH13ANALYSIS

OF

THE

REDIRECTION

GRAPH

GFrom

now

on,

the

resultsareonly

about

the

old

graph.14LONG

REDIRECTION

PATHS[‘/resource/Mirage_%28pop_group%29','/resource/Mirage_(pop_group)','/resource/Mirage_(pop_group)','/page/Mirage_(pop_group)','/page/Mirage_(pop_group)','/resource/Mirage_(disambiguation)','/resource/Mirage_(disambiguation)','/page/Mirage_(disambiguation)',‘/page/Mirage_(disambiguation)']1545.1%

(encoding,

http->https,

upper/lower

case)16.8%

(DBpedia

resource

to

page,

etc.)Approx.

45.1%

-

83.2%

canbe

taken

as

identity

linksIMPLICIT

SEMANTICS

OF

REDIRECTION

(4,000

EDGES)16100

CHAINS

OF

REDIRECTIONOn

average1.7hops.

We

examine

redirection

chains

with

over

2

hopsSimilar

number

of

hops

for

RUF,

RUE,

RUNF,

RUT.

So

we

sample

uniformly85%

happens

within

a

domainWikidata

(28%)andDBpedia

(25%)

are

among

the

most

observedChains

of

DBpedia

are

often

among

the

longest.Some

others

were

observed

for

(5%)

and

(1%)17•18Few

entities

can

be

redirected

from

G

to

HRedirection

is

a

well-observed

in

identity

graphsWhen

only

1-3%

can

bedereferenced,

it

hurts

accessibility

and

interoperabilityWe

observ