Group 1: 5th presentation
Comparing academic hyperlink structure with co-authorship pattern in Korea
Hyo Kim@ Ajou University
Han Woo Park@ YeungNam University
Hyo Kim
College of Information TechnologyMedia divisionAjou UniversityKorea (South)
Tel) +82-31-219-1858E-mail) [email protected]
Han Woo Park
School of Social SciencesYeung Nam University
Korea (South)[email protected]
http://www.hanpark.net
Study
• Structural characteristics of academic hyperlinks among universities in Korea
• Relationship between hyperlinks and productiveness
• Speculation of actual communication patterns from the hyperlink activities
• Via SNA (social network analysis) approach
Data I
• http://www.braintrack.com/– The most visible two universities in each local
region in Korea (South part, N = 30)• http://altavista.com/
– Number of in-links and out-links to the universities in the data set
• ISI database – the number of research articles listed in the S
CI index published in each university– 2 univ. dropped (N = 28)
Data II – # of hyperlinks
Universities - - - - - -
snu 139 86 31 67 …
korea 90 14 10 16 …
pusan 95 26 14 25 …
donga 20 8 9 9 …
kyungpook 36 13 11 1 …
keimyung 4 3 1 0 0 …
hannam 10 5 2 2 2 …
… … … … … … …
Data III
Indegree outdegree
snu 1224 1070
korea 324 471
pusan 393 373
donga 158 171
kyungpook 152 371
keimyung 27 137
inchon 56 69
chonnam 1088 287
chosun 104 182
… … …
Dichotomization for CONCOR
• From the initial matrix (Data II)• Replacing binary values
– Average of the matrix = 17.11– Cells bellow the mean = 0– Cells greater than or equal to (GE) the
mean = 1
• New data matrix (see next page)
CONCOR - dichotomized
snu korea pusan dongaKyungpook …
snu 0 1 1 1 1 …
korea 1 0 0 0 0 …
pusan 1 1 0 0 1 …
donga 1 0 0 0 0 …
Kyungpook 1 0 0 0 0 …
… … … … … … …
Groups identified by CONCOR
MDS graph
GroupsIdentified(from CONCOR)
10-1
1
0
-1
snu
korea
pusan
donga
kyungpook
keimyung
hannam
cnu
inha
inchon
chonnam
chosun
cheju
hallym
kangwon
ajou
suwon
chungbuk
hoseo
schyeungnam
cataegu
gsnu
changwon
chonbukwonkwang
mokpo
daebul
A
B
C
D
10-1
1
0
-1
snu
korea
pusan
donga
kyungpook
keimyung
hannam
cnu
inha
inchon
chonnam
chosun
cheju
hallym
kangwon
ajou
suwon
chungbuk
hoseo
schyeungnam
cataegu
gsnu
changwon
chonbukwonkwang
mokpo
daebul
10-1
1
0
-1
snu
korea
pusan
donga
kyungpook
keimyung
hannam
cnu
inha
inchon
chonnam
chosun
cheju
hallym
kangwon
ajou
suwon
chungbuk
hoseo
schyeungnam
cataegu
gsnu
changwon
chonbukwonkwang
mokpo
daebul
A
B
C
D
Relationships among the groups
A B C D
A 1 0 0.93 0.91
B 0 0.33 0 0
C 0.79 0 0.53 0.089
D 0.31 0 0 0.06
Average 0.31
Visualization of group rel.
• Group A, B, C, D (identified from CONCOR) can be visualized
.31
.53
.79
.91
1
.93
AB
CD
Group rel.• Members in group A = the strongest rel (regardin
g hyperlinks, value = 1)• Members in group C = strong rel (value = .53)• Strong rel between group A and C (A->C = .93; C
-> A = .79)• Members in group B = isolated• Members in group D = no strong hyperlink activity
among themselves, but, strong hyperlink-receivers (value = .91) and weak hyperlink maker (value = .31)
ANOVA
• Are these groups meaningful in terms of the number of links (in and out); and the number of articles?– In-links: F (3, 24) = 9.73, p < .0001– Out-links: F (3, 24) = 62.79, p < .0001– Articles: F (3, 24) = 8.26, p < .0001
• Group A differs from all other universities in terms of the number of published journal articles, which means the members of group A are strong research universities.
QAP (dyadic rel)
• CONCOR test just reveals general relationships among members in each group or among groups.
• ANOVA test does not reveal relationships.
• QAP will reveal specific relationship between universities and SCI articles at a dyadic level.
QAP test
• IV: in- and out-links matrices• DV: matrix of the number of articles
DV = SCI journal articles
sci
snu 3828
korea 1193
pusan 757
donga 191
kyungpook 951
keimyung 159
hannam 62
cnu 635
inha 717
• The number of SCI journal articles
• The data set is not usable for QAP test because it is an attribute data (just one raw, it has).
• So, the data is transformed into matrix (via obtaining the dyadic difference of the number of articles between two universities)
DV = SCI journal articles
snu korea pusan donga …
snu 0 2635 3071 3637 …
korea 2635 0 436 1002 …
pusan 3071 436 0 566 …
donga 3637 1002 566 0 …
… … … … … …
• Each number in a cell means the absolute difference (of the number of articles) between two universities
QAP result beta P p low p high
Intercept
0 0.958 0.958 0.042
OUT 0.24 0.007 0.007 0.993
IN 0.41 0.04 0.04 0.96
• R-square = 35.6%
• Both In and Out matrix are significantly related to the DV matrix (# of articles; In = .41; Out = .24), which means . . .
• at a dyadic level, if one university has more links (both in and out), the university produces more SCI journal articles.
• Caution: a kind of regression test, which means
• no causal relationship between IVs and DV are assumed.
• Therefore, we can just speculated that SCI journals are significantly related to number of links.
Study discussion
• With SNA, we explored• At structural level
– The structural characteristic of the whole matrix of in and out hyperlinks.
– Four groups identified from the structural characteristics
– Four groups differed from # of SCI articles, which means hyperlinking activity is related to the journal publication.
Study Discussion II
• At a dyadic level,– Specifically, # of SCI articles is related
to # of in and out links between two universities.