Information Technology Reference
In-Depth Information
1
0.8
0.6
Train
Test
0.4
0.2
0
Fig. 2
Evaluation for each attribute-based ranking as well as centrality-based ranking with
target ranking among researchers
1
0.8
0.6
Train
Test
0.4
0.2
0
Fig. 3
Evaluation for network rankings in a combined-relational network with
Paper
among
researchers
web sites
G
Jcooc
and a co-affiliation network
G
af filiation
,and
Masaru Kitsuregawa
has stable centralities on several networks.
For the baseline model, three centrality indices (degree centrality
C
d
, closeness
centrality
C
c
, and betweenness centrality
C
b
) are used on different networks (
G
Ecooc
,
G
Eoverla p
,
G
Jcooc
,
G
Joverla p
,
G
af filiation
,and
G
pro ject
) as network rankings. We cal-
culate the correlation between network rankings with each target ranking of
Pa-
per
. For comparison, we also rank companies according to previously described
attributes (i.e.,
JhitNum
and
EhitNum
), and take correlation with target ranking.
Actually, Fig. 2 portrays correlations (mean of three tries) of each network rank-
ing as well as each attribute-based ranking with different target rankings on training
and testing data among researchers. Results show that the hit number of names on
Japanese web sites is a good attribute of researchers for predicting the creditability
of publications. Furthermore, degree centralities in an overlap network, as they do
in a cooc network on English-language web sites (
r
G
Eoverla p
,
C
d
and
r
G
Ecooc
,
Cd
) exhibit