Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
----An example of wetland remote sensing
Building semantic knowledge organization systems for interdisciplinary research
Asian NKOS workshop, Seoul,Dec,9th,2015
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
Outline
l 1. Introduction: background, aim l 2. Experimental design: workflow
l 3. Results and discussions l 4. Conclusions and future work
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉ STKOS ontology DB
STKOS metathesaurus Sharing system
1. Introduction l Background
?
Interdisciplinary and emerging field:
• brain Science • quantum
communications • wetland
How to use
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
1. Introduction l Aims
developthemethodsandworkflowsinbuildingseman6cknowledgeorganiza6onsystem(KOS),insupportofinterdisciplinaryscien6ficresearch
l Experimental Field – Wetland
• remotesensing• Biosphere • Atmosphere • Lithosphere • ……
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
documents Doc
processing Terms &
frequency
Mapping to STKOS concepts STKOS
metathesaurus
Selecting concept
Analyze semantic type & relations
KOS
indexing
query Query expansion
List of concepts, relations
Search and ranking
2. Workflow
Model Construction
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
2. Workflow
l Data source – TermsfromcurrentavailableKOS
• STKOS metathesaurus: 199 KOS ( 150 thesauri, 3 classifications, 37 glossaries, etc.)
• Remote sensing lexicon, Science publishing, 1990
– Termsextractedfromliterature• Web of Science Database(1900-2014), Topic field
= “remote sensing & wetland”, bibliographic records
l Methods – Seman6cmatching– Domainexpertspar6cipa6on
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
documents
Doc processing
terms、frequency
Mapping to STKOS concepts STKOS
metathesaurus
2. Workflow
l Pre-processing / cleaning: Thomson Data Analyzer(TDA) – Stopwordremoval,analysis,clustering
l Semantic match to STKOS: – Normaliza6on,phraseanalysis,wordsenseanalysis,structureanalysis– termlist
l term extraction – Termsselectedoveraspecificthreshold:termfrequency(TF)Thresholdterm=mean(TF+standarddevia6onTF),wherehighTF,mediumTF,,lowTF
Knowledge collecting
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
2. workflow
l domain expert participation – Reconstructandreusemodel,Remotesensinglexicon– Removegeneralterms,suchas“classifica6on,casestudy”– Selectcompoundterm– Category,Seedterm,doubly-anchoredpaSerns
Select concept
Analyze semantic type & relations
KOS Construct model
Construct KOS model
Doc processing
terms、frequency
Mapping to STKOS concepts STKOS
metathesaurus
Iteration
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
documents indexing
Domain expert evalution
2. Workflow Constructed
KOS
l Evaluation – Annota6onandindexbyhumanandcomputer– Precision,recall,howcorrectistheKOSlearned
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3. Results and discussions
frequency term MatchSTKOS percent Match+Standarddeviation percent
1 4271 707 14.98% 707 14.98%
2 548 178 32.48% 178 32.48%
3 196 85 43.37% 87 44.39%
4 106 57 53.77% 58 54.72%
5 65 41 63.07% 43 66.15%
6 40 21 52.5% 22 55%
7 28 19 67.86% 19 67.86%
8 25 17 68% 18 72%
9 19 14 73.68% 14 73.68%
10 11 5 45.45% 5 45.45%
11-20 47 30 63.83% 31 65.96%
21-30 9 9 100% 9 100%
31-40 8 7 87.5% 7 87.5%
41-50 6 5 83.3.% 6 100%
> 50 5 5 100% 5 100%
total 5834 1200 20.57% 1209 20.72%
l Match STKOS result
Threshold term =4
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3. Results and discussions l Source of STKOS matching result
508 401
364 357
336 331
297 266
263 260
252 241
237 217 196 195
192 162
135 132
129 127
125 109
108 95 88
84 83
83 79
77 68
60 54
50 43
39 29 19
14 13 7 5 3 1 1
GE
OR
EF
THE
SA
UR
US
9TH
N
AS
A TH
ES
AU
RU
S
THE
SA
UR
US
OF
EM
TRE
E T
HE
SA
UR
US
E
DTE
INIS
JO
INT
AN
TE_A
BS
TRA
CTS
IN N
EW
B
IOC
OM
PLE
XIT
Y U
NE
SC
O T
HE
SA
UR
US
LI
FE S
CIE
NC
ES
TH
ES
AU
RU
S
GE
NE
RA
L M
ULT
ILIN
GU
AL
EU
RO
VO
C (M
ULT
ILIN
GU
AL
FLO
RID
A E
NV
IRO
NM
EN
TS
THE
SA
UR
US
ETH
ICS
IN T
HE
W
ATE
R R
ES
OU
RC
ES
IN
SP
EC
TH
ES
AU
RU
S
EI T
HE
SA
UR
US
20
09 IE
EE
TH
ES
AU
RU
S
SE
AR
S L
IST
OF
SU
BJE
CT
EV
ER
GLA
DE
S O
NLI
NE
C
A G
EN
ER
AL
SU
BJE
CT
WQ
PB
LIB
RA
RY
THE
SA
UR
US
P
OLL
UTI
ON
TH
ES
AU
RU
S
US
GS
TH
ES
AU
RU
S
EU
RO
PE
AN
TR
AIN
ING
G
LOB
AL
CH
AN
GE
MA
STE
R
IVO
AT T
HE
SA
UR
US
E
NV
IRO
NM
EN
T, S
AFE
TY A
ND
E
.V. K
OM
AR
EK
FIR
E
ED
UC
ATIO
NA
L R
ES
EA
RC
H
THE
SA
UR
US
OF
THE
AS
TRO
NO
MY
THE
SA
UR
US
.MAT
HS
.OR
G
CO
PP
ER
TH
ES
AU
RU
S
THE
INTE
GR
ATE
D P
UB
LIC
A
DR
AFT
EN
VIR
ON
ME
NTA
L U
SG
S G
ATE
WAY
TH
ES
AU
RU
S
DE
PAR
TME
NT
OF
THE
S
SD
TH
ES
AU
RU
S
BIO
ETH
ICS
TH
ES
AU
RU
S
ALE
XA
ND
RIA
DIG
ITA
L M
ATH
TH
ES
AU
RU
S
THE
MAT
H F
OR
UM
INTE
RN
ET
BR
ITIS
H_M
US
EU
M
AG
RIC
OLA
TH
ES
AU
RU
S F
OR
TH
ES
AU
RU
S O
F C
GI S
IMP
LE L
ITH
OLO
GY
SK
Y TH
ES
AU
RU
S
THE
SA
UR
US
FO
R T
HE
HE
RB
source of matching stkos
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3. Results and discussions
Wetland remote sensing
General
RS classification
Resolution
RS base
Physics
Math
Geoscience
Information science
RS technology
Remote Sensor
platform
ground station
image processing
Photography
Interpretation
Wetland RS application
KOS model of Wetland remote sensing
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3 Results and discussions
l Semantic KOS of Wetland RS – Classifica6on:4primary,12secondary– Concept:209– Term:409– Rela6onship:hierarchical,associa6ve,equivalence
– Seman6ctypeandrela6on:equipment,product,concept?
Rela6on:EquipmentAisakindofB
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3. Results and discussions ParentTerm• satellite
PreferredTermEnvironmentalSatelliteNon-preferredTermEarthobserva6onsatelliteENVISATChinese: 环境卫星Seman6cType:Equipment
ChildrenTerms• IKONOS• QuickBird• RapidEye• SPOT• WorldView• GeoEye• Landsatsatellites
1
2
3
Equipment • Sensors • Thematic mapper • Imaging spectrometer
Product • Image • Data
Application • Land hydrology • Ecological community • Crop
Sem
antic
re
latio
nshi
ps
4
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
3.Results and discussions
Computer Article Number
Evaluation by domain experts
Index term Number
Correct Error
0 81 31 50
1 102 79 23
2 93 83 10
3 110 97 13
4 135 126 9
5 126 124 2
6 150 148 2
>7 765 765 0
Evaluation Index term, index term number, index semantic type • Does index term represent the content in wetland RS • Need to add some terms • Is the article belong to wetland RS?
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
4. Conclusions and future work
l Conclusions – MethodsofinterdisciplinaryKOSconstruc6on– Coverthecontentofwetlandremotesensing– UseinwetlandDB– Improveretrievalefficiencyanddiscover
l Future work – Seman6ctyperevise– con6nueinotherfieldofwetland
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
Our Team
Liu Zheng1, Sun Tan2, Sheng Chunlei3, Liu Xiumin1, 1 National Science Library, Chinese Academy of Sciences
2 Agricultural information Institute, Chinese Academy of Agricultural Sciences
3 Northeast institute of Geography and Agroecology, Chinese Academy of
Sciences
Wetland experts Lv Xiangguo, Song Kaishan, Wang Zongming
Key laboratory of wetland ecology and environment, Northeast institute of
Geography and Agroecology, CAS
学科馆员: ⻘秀玲
报告时间: 2010 年 4 ⽉
Thanks Any questions ?