AGE-DECAYING H-INDEX FOR SOCIAL NETWORKS OF CITATIONS
Presentation by:
Apostolos Papadopoulos
2D. Katsaros
1,2, A. Sidiropoulos
2, Y. Manolopoulos
21
Dept. of Computer & Communication Engineering University of Thessaly, Volos, Greece
2
Dept. of Informatics
Aristotle University, Thessaloniki, Greece
Methods for Ranking Scientists
z Evaluation of scientists by “experts”
–
e.g., surveys
z Citation Analysis
–
Task: Compute a score for the “objects”
z Hybrid method of previous two
Drawbacks of various scientists ranking methods
z Not measure the importance of papers
z Affected by “big hits”
z Not measure productivity
z Need to set administrative parameters
H-index
z Published by J.E. Hirsch in Oct. 2005
z Was known at least two years earlier
z Definition:
A researcher has h-index h if
–
h of his N
particles have received at least h citations each
–
the rest N
p-h articles have received no more than
h citations each
H-index
z Calculates the broadness of a researcher
9
Productivity
9
Impact
9
Not affected by “big hits”
9
Not affected by “noise”
Index a
z N
c,tot≥h
2z Definition:
A researcher has index a if N
c,tot=ah
2z Second metric-index
H-index drawbacks
z It is a growing function over time
z Does not show scientist’s inactivity or retirement
z Scientists with short scientific life are out of
competition
Contemporary H-index
z
Definition
(in Scientometrics, 72(2), 2007):
A researcher has contemporary h-index h
cif
– hc
of his N
particles have S
c(i)≥h
c–
the rest N
p-h
carticles have S
c(i)≤h
c– Sc(i)=
γ
* (Y(now) - Y(i) + 1)-δ |C(i)|–
In our experiments: γ=4 and δ=1
z
An old article gradually loses its “value”
z
Show how “active” a researcher is
Trend H-index
z
Definition
(in Scientometrics, 72(2), 2007): A researcher has trend h-index h
tif
– ht
of his N
particles have S
t(i)≥h
–
the rest N
p-h
tarticles have S
t(i)≤h
–
In our experiments: γ=4 and δ=1
z
An old citation gradually loses its “value”
z
Shows how “trend” the work of a researcher is
δ
−
∈
∀
+
−
= γ * ∑ ( Y ( now ) Y ( i ) 1 )
(i) S
C(i) x t
Age Decaying H-index
z
Definition:
A researcher has Age Decaying h-index h
adif
– had
of his N
particles have S
ad(i)≥h
ad–
the rest N
p-h
adarticles have S
ad(i)≤h
ad–
In our experiments: γ=4 and δ
1=δ
2=1
z
An old article gradually loses its “value”
2
1
* ( ( ) ( ) 1 )
) 1 )
( )
( (
2
*
δ −δ∈
∀
−
− +
+
−
= γ Y now Y i ∑ Y now Y i
(i) S
C(i) x
ad
The second metric “a” of h-index
z
A factor a is defined as:
– Nc,tot=ah2
z
For h-index generalizations, instead of N
c,totwe use the sum of each paper score S
ad(i):
2 ad ad
P i
ad
(i) a h
S =
∑
∈∀
H-index generalizations
z Contemporary
z Trend
z Age Decaying
–
Scientists, journals, conferences or any other kind
of semantic grouping of articles
Experiments
z DBLP collection (http://dblp.uni-trier.de/)
–
Data timestamp: March 2006
z DBLP includes data for authors, journals and conferences
z Focuses in the DB area
z “Names Problem” is solved
Experiments – h-index
130 913
2.81 18
8.Umeshwar Dayal
154 994
3.06 18
7.Rakesh Agrawal
93 1024
3.16 18
6.Catriel Beeri
143 1071
2.96 19
5.Won Kim
124 1359
3.39 20
4.Philip A. Bernstein
150 1896
3.91 22
3.David J. DeWitt
227 1783
3.37 23
2.Jeffrey D. Ullman
193 2180
3.78 24
1.Michael Stonebraker
Np
Nc,tot
a h
Name
Experiments – Age Decaying h-index
1 1
3 3.41
14 7.David J. DeWitt
2 2
2 4.18
14 6.Jeffrey D. Ullman
13 21
69 2.85
15 5.Alon Y. Levy
25 22
100 2.79
16 4.Dan Suciu
6 5
13 3.08
16 3.Serge Abiteboul
5 6
23 3.19
16 2.Jennifer Widom
4 4
7 3.28
16 1.Rakesh Agrawal
Pos by ht Pos
by hc Pos
by h aad
had Name
Experiments – h-index for scientists
Experiments – h-index for scientists
Experiments – h-index for scientists
Experiments – H-index for conferences
42 412
2.86 12
8.adbt
101 527
3.65 12
7.eds
434 658
3.89 13
6.edbt
1338 1486
5.80 16
5.er
1970 3307
6.83 22
4.icde
776 3883
5.74 26
3.pods
2192 9729
7.10 37
2.vldb
2059 12261
6.05 45
1.sigmod
Np Nc,tot
a h Name
3 163
31 2.69
9 8.webdb
13 434
658 3.92
11 7.edbt
6 1074
243 4.08
11 6.kdd
11 313
580 5.27
12 5.icdt
22 1970
3307 8.01
17 4.icde
26 776
3883 5.32
20 3.pods
37 2192
9729 6.94
25 2.vldb
45 2059
12261 5.85
32 1.sigmod
Np h
Nc,tot ac
had Name
Experiments – H-index for conferences
Experiments – H-index for conferences
Experiments – H-index for journals
4939 388
6.06 8
8.ipl
281 408
5.03 9
7.vldb
877 863
7.13 11
6.debu
378 740
4.37 13
5.tois
1349 1142
5.07 15
4.sigmod
934 1208
4.71 16
3.is
1388 1520
4.69 18
2.tkde
598 9329
3.88 49
1.tods
Np Nc,tot
a h Name
6 318
156 5.67
6 8.jiis
16 934 1208
7.51 6
7.is
6 238
189 3.82
7 6.dpd
9 281
408 2.82
12 5.vldb
11 877 863
3.49 12
4.debu
18 1388
1520 5.77
12 3.tkde
15 1349
1142 4.94
13 2.sigmod
49 598 9329
7.71 13
1.tods
h Np
Nc,tot aad
had Name
Experiments – H-index for journals
Conclusion
z
Evaluation of scientists based on citation analysis
z
Evaluation of publication forums based on citation analysis
z
H-index shortcomings:
–
Active – inactive scientists
–
Significant works in the past – not any more significant
zH-index generalizations on the time dimension
AGE DECAYING H-INDEX FOR SOCIAL NETWORKS OF CITATIONS
Thank you for your attention!
D. Katsaros, A. Sidiropoulos, Y. Manolopoulos
Presenter: Apostolos Papadopoulos