Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
LEIDEN UNIVERSITY 2012
1. Advanced bibliometric methods for
ranking and benchmarking of
universities
Middle East Technical University
Ankara, October 15, 2012
Ton van Raan
Center for Science and Technology Studies (CWTS)
Leiden University
2.
3.
4. Leiden University
oldest in the Netherlands, 1575
European League of Research Universities (LERU)
Within world top-100 in all international
university rankings
12 Nobel Prizes
Leiden, historic city (2th, 11th C.), strong cultural
(arts, painting) & scientific tradition
one of the largest science parks in EU
5. Contents of this presentation:
* Bibliometric methodology & practical
applications:
• impact
• maps
* Leiden Ranking 2011-2012: new indicators
* Evaluation tools related to the Leiden Ranking
6. Total publ universe
non-WoS publ:
Books
Book chapters
Conf. proc.
Reports
WoS/Scopus sub-universe
journal articles only,> 1,000,000p/y Non-WoS journals
VOLUME 88, Number 13 PHYSICAL REVIEW LETTERS 1 April 2002
VOLUME 88, Number 13 PHYSICAL REVIEW LETTERS Behavior in “Scale-Free” Network Models
Truncation of Power Law 1 April 2002
due to Information Filtering
Truncation of Power Law Behavior in “Scale-Free” Network ModelsVOLUME 88, Number 13 PHYSICAL REVIEW LETTERS 1 April 2002
due to Information Filtering Barthélémy,3 H. Eugene Stanley,1 and Luís A. Nunes Amaral1
Stefano Mossa,1,2 Marc
VOLUME 88, Number 13 PHYSICAL REVIEW LETTERS 1 Center forApril 2002 and Department of Physics, Boston University, Boston, Massachusetts 02215
1 Polymer Studies Truncation of Power Law Behavior in “Scale-Free” Network Models
Stefano Mossa,1,2 Marc Barthélémy,3 H. Eugene Stanley,1 and LuísUdR, and INFM Center for Statistical Mechanics and Complexity, Information Filtering
2 Dipartimento di Fisica, INFM A. Nunes Amaral1 due to
Truncation of Power Law1Behavior in “Scale-Free” Network Models di Roma “La Sapienza,” Piazzale Aldo Moro 2, I-00185, Roma, Italy
Università
Center for Polymer Studies and Department of Physics, Boston University, Boston, Massachusetts 02215
2 Dipartimento di Fisica, INFM UdR, and3 CEA-Servicefor Statisticalde la Matière Condensée, BP 12, 91680 Bruyères-le-Châtel, France
due to Information Filtering INFM Center de Physique Mechanics and Complexity, Stefano Mossa,1,2 Marc Barthélémy,3 H. Eugene Stanley,1 and Luís A. Nunes Amaral1
(Received 18 October 2001; published 14 March 2002)
Università di Roma “La Sapienza,” Piazzale Aldo Moro 2, I-00185, Roma, Italy 1 Center for Polymer Studies and Department of Physics, Boston University, Boston, Massachusetts 02215
Stefano Mossa,1,2 Marc Barthélémy,3CEA-Service de Physique Luís A.We formulate a general 91680 Bruyères-le-Châtel, of scale-free Dipartimento di Fisica, INFM UdR, and INFM Center for Statistical Mechanics and Complexity,
3 H. Eugene Stanley,1 and de la Matière Condensée, BP 12,
Nunes Amaral1 France 2
(Received Boston, Massachusetts 02215 model 2002)the growth
18 October 2001; published 14 March for networks under filtering information
1 Center for Polymer Studies and Department of Physics, Boston University, conditions—that is, when the nodes can process information about only a subset of the di Roma “La Sapienza,” Piazzale Aldo Moro 2, I-00185, Roma, Italy
Università
existing nodes in the
2 Dipartimento di Fisica, INFM UdR, and INFM Center for Statistical Mechanics and Complexity, distribution of the number of incoming linksCEA-Service de Physique de la Matière Condensée, BP 12, 91680 Bruyères-le-Châtel, France
3
We formulate a general model for the growth that the networks under filtering information to a node follows a universal scaling
network. We find
Università di Roma “La Sapienza,” Piazzale Aldo Moro 2, I-00185, Roma,that itof scale-freepower law with an exponential truncation controlled not only(Received 18 October 2001; published 14 March 2002)
Italy decays
conditions—that is, when 12, 91680 Bruyères-le-Châtel, France as aonly a
form, i.e.,information about
3 CEA-Service de Physique de la Matière Condensée, BP the nodes canbut also by a feature not previouslysubset of thethe subsetnodes in the “accessible” to the node. We test our
process existing by the system size
considered, of the network
(Received network. We find published 14 March 2002) number of incoming for theto a node follows aand find agreement.
18 October 2001; that the distribution of the with empirical data links World Wide universal scaling
We formulate a general model for the growth of scale-free networks under filtering information
form, i.e., that it decays as a power law modelan exponential truncation controlled not Web by the system size
with only conditions—that is, when the nodes can process information about only a subset of the existing nodes in the
We formulate a general model for the growth of scale-free networks under the subset of the network “accessible” to the node. We test ourWe find that the distribution of the number of incoming links to a node follows a universal scaling
but also by a feature not previously considered, filtering information network.
DOI: 10.1103/PhysRevLett.88.138701 PACS numbers: 89.20.Hh, 84.35.+i, 89.75.Da, 89.75.Hc
conditions—that is, when the nodes canmodel with empirical data for the World Widethe existing nodes in the
process information about only a subset of Web and find agreement. form, i.e., that it decays as a power law with an exponential truncation controlled not only by the system size
network. We find that the distribution of the number of incoming links to a node follows a universal scaling but also by a feature not previously considered, the subset of the network “accessible” to the node. We test our
form, i.e., that it decays as a power law with an exponential truncation controlled not onlynumbers: 89.20.Hh, 84.35.+i, 89.75.Da, 89.75.Hc model with empirical data for the World Wide Web and find agreement.
DOI: 10.1103/PhysRevLett.88.138701 PACS by the system size
but also by a feature not previously considered, the subset of theThere is a“accessible” to the node. We in understanding the structure and growth mechanisms of global networks [1–3], such as the World Wide
network great deal of current interest test our
model with empirical data for the World Wide Web and find agreement. DOI: 10.1103/PhysRevLett.88.138701 PACS numbers: 89.20.Hh, 84.35.+i, 89.75.Da, 89.75.Hc
Web (WWW) [4,5] and the Internet [6]. Network structure is critical in many contexts such as Internet attacks [2], spread of an Email virus [7], or
There is a great deal of current interest in understanding the structure and growththese problems, the nodes with the largest number of links Widean important role on the dynamics of the
dynamics of human epidemics [8]. In all mechanisms of global networks [1–3], such as the World play
DOI: 10.1103/PhysRevLett.88.138701 PACS numbers: 89.20.Hh, 84.35.+i, 89.75.Da, 89.75.Hc the global structure of the network as well as its precise distribution of the number of links.
Web (WWW) [4,5] and the Internet [6]. NetworkItstructure is critical in many contexts such as Internet attacks [2], spread of an Email virus [7], or
system. is therefore important to know
dynamics of human epidemics [8]. In all Recent empiricalthe nodesreportthe largest the Internetlinks playWWW have scale-free properties; that understanding the structure and growththe
these problems, studies with that both number of and the an important deal on current interestofinthe the number of incoming links and mechanisms of global networks [1–3], such as the World Wide
There is a great role of the dynamics
is,
system. It is therefore important to know number of structure of the at a given node as its precise distribution of the number ofthe Internet [6]. Network structure is critical in the scale-free such as Internet attacks [2], spread of an Email virus [7], or
the global outgoing links network as well have distributionsWeb (WWW) [4,5] and law tails [4–6]. It has been proposed [9] that many contexts
that decay with power links.
Recent empirical studies report growthstructure of the InternetnetworksWWW may beproperties; bydynamics of human epidemics links all these problems, in which new nodes link
the Internet and the and the [1–3], scale-free explained Widemechanism referred to as “preferential attachment” [10]
a
There is a great deal of current interest in understanding the structure andthat bothmechanisms of globalWWW have proportionalthe Worldthat is, the number of to these nodes.Inand the focus on thethe nodes with the largest number of links play an important role on the dynamics of the
incoming [8].
to existing nodes with a decay with such as to [4–6]. It
number structure is links at a given node have distributions thatprobability power Emailthe number of existing links important to know the Here we stochastic character of the
Web (WWW) [4,5] and the Internet [6]. Networkof outgoing critical in many contexts such asattachment mechanism, which anlaw tailsvirus in thehas been proposed [9]nodesthe scale-free globalthe existing nodes with the largest its precise distribution of the number of links.
system. It is therefore that
preferential Internet attacks [2], spread of we understand [7], orfollowing way: in which want to connect to
structure of the network as well as
attachment” [10] New
dynamics of human epidemics [8]. In all structure of the Internet and the the largest number of links by a withimportant role onto asdynamics thethe empirical studies report that to alink Internet and the For a large network it properties; that is, the number of incoming links and the
these problems, the nodes with WWW may be explained playmechanism referred the “preferential advantages offered by beingnew nodes well-connected node. WWW have scale-free
Recent both the
number of links—i.e., of the largest degree—because of of we focus
an linked of the
system. It is therefore important to knowto existing structure of a probabilityas well as its precisenumbernewexistingnumber of links. of all number ofon the stochastic charactermake a decision on which node to connect with law tails [4–6]. It has been proposed [9] that the scale-free
the global nodes with the network proportional to the distributionnode will know these nodes. Here existing nodes, so a new node given node have distributions that decay with power
not plausible that a of the links to the degrees outgoing links at a
is we understand in the following way: New nodes want to connect to the existing nodes with the largest
must
Recent empirical studies report that both the Internet and the WWW have which properties; that is, theit number of the state of the and structure of the Internet and the WWW may be explained by a play as nodes with ato as “preferential attachment” [10] in which new nodes link
preferential attachment mechanism, scale-free
has about incoming links network.
the mechanism referred
number distributions with the largest degree—because informationbeen offered [9] that the scale-free The preferential attachment mechanism then comes into
based on what of the
number of outgoing links at a given node have of links—i.e.,that decay with power law tails [4–6]. Itadvantagesproposedby being linked to ato existing nodesnode. For a large network it to the number of existing links to these nodes. Here we focus on the stochastic character of the
well-connected with a probability proportional
larger degree of more has to become
is not plausible that mechanism referred to degreesare all existing nodes, so known. new nodes link
likely
structure of the Internet and the WWW may be explained byaanew node will know the as “preferential attachment” [10] ainnew node must make a decisionattachmentnode to connect with understand in the following way: New nodes want to connect to the existing nodes with the largest
which preferential on which mechanism, which we
Refs > non-WoS
to existing nodes with a probability proportional to the number of existingabout to these nodes. Here we focus preferential attachment mechanism then links—i.e., with the nodes with a
based on what information it has links the state of the network. The on the stochastic character of number of comes into play as largest degree—because of the advantages offered by being linked to a well-connected node. For a large network it
the
preferential attachment mechanism, which we understand more likely to become New nodes want to connect to the existing nodes with the largest plausible that a new node will know the degrees of all existing nodes, so a new node must make a decision on which node to connect with
larger degree are in the following way: known. is not
number of links—i.e., with the largest degree—because of the advantages offered by being linked to a well-connected node. For a large network it on what information it has about the state of the network. The preferential attachment mechanism then comes into play as nodes with a
based
is not plausible that a new node will know the degrees of all existing nodes, so a new node must make a decision on which node to connect with degree are more likely to become known.
larger
based on what information it has about the state of the network. The preferential attachment mechanism then comes into play as nodes with a
larger degree are more likely to become known.
6
8. LUMC Research Program Neurosciences
Neuro
Radiology &
Brain Imaging
Neurology Diseases of the
brain
Brain
Surgery Genetics of neuro diseases
Human Epidemiology & Medical
Genetics Statistics
Degree coefficient correlation 8
negative
9. Radiology, Nuclear Medicine, & Medical Imaging
111 journals, 2007-2010, Citation-based network map
Colors indicate the different subfields
15. Neurology 2005-2009, CD Map, n=2,000
Normalization based on ‘formal’ field definition:
WoS field (journal category) Neurology
Large differences between clinical and basic science part
Clinical research has disadvantage by
normalization with average field citation
density
CWTS is developing a new normalization
procedure to solve this problem!
15
17. 1992 - 2000 PUBLICATIONS AND IMPACT PER SUBFIELD
Diseases of the Neurosystem - 2000
1992
Dept Output in Fields
Erkrankungen des Nervensystems Dept Impact from Fields
Erkrankungen des Nervensystems
CPP/
SUBFIELD
FCSm
(CPP/FCSm)
Neurosc
NEUROSCIENCES 2.28
(2.28)
Biochem & MOL BIOL 2.00
BIOCH (2.00)
CPP/
SUBFIELD
(CPP/FCSm) Cell Biol BIOLOGY 1.96
CELL (1.96)
FCSm
Devel Biol
DEVELOPMENT BIOL 1.33
(1.33)
Neurosc 2.70
NEUROSCIENCES (2.70)
multidisc
Knowledge use by very
MULTIDISCIPL SC 2.84
(2.84)
Genetics & HERED 2.70
GENETICS (2.70)
Biochem 1.92
BIOCH & MOL BIOL (1.92)
Physiol 1.74
PHYSIOLOGY (1.74)
high impact groups
Cell Biol (2.74)
CELL BIOLOGY
2.74 Clin Neurol
CLIN NEUROLOGY 1.58
(1.58)
Pharmacol & PHAR 3.94
PHARMACOL (3.94)
multidisc SC (4.33)
MULTIDISCIPL 4.33 OncologyONCOLOGY 1.60
(1.60)
Biology 2.06
BIOLOGY (2.06)
Devel Biol 1.32
EVELOPMENT BIOL (1.32) Anat ANAT & MORPHOL 2.15
Morph (2.15)
0% 10% 20% 30% 40% 50%
Endocrinol & METAB 1.81
60%
ENDOCRIN (1.81) Societal
Percentage of Publications Pathology
PATHOLOGY 2.11
(2.11) impact:
Urology & NEPHRO 1.30
UROLOGY (1.30)
Basic science
Biophysics
BIOPHYSICS 1.97
used in applied
(1.97)
Immunol
IMMUNOLOGY 1.84
(1.84)
Res Medic
MEDICINE, RES 1.88
fields
Reassignment by field,
IMPACT: LOW AVERAGE HIGH
Zoology
(1.88)
4.51
Will work next month
ZOOLOGY (4.51)
Opthalmol
OPHTHALMOLOGY 2.21
(2.21)
0% 5% 10% 15% 20% 25% 30% 35% 40%
Relative share of impact
18. CWTS online report
• Web-based system for user-friendly exploration of
CWTS research performance studies
• Supported by various types of visualizations
• Bibliometric indicators at various aggregation levels
e.g., from university as a whole to specific research
groups
• Flexibility in the choice of indicators and time periods
• Design is based on over 25 year experience
21. On the problem of ranking
• Rankings have become ‘indispensable’
• Methodological problems have political consequences
• Key deficiencies:
– Reduction multi-dimensional to 1-dimensional list
– Lack of transparency of most global rankings
– Data quality often insufficient
– Bad fit between ranking assumptions and diversity
university missions
22. Indicators divided into 5 categories, with fraction of final
ranking score
Teaching: 30 %
Research: 30 %
Citations: 30 %
Industry income: 0.25 %
International: 0.75 %
27. Leiden Ranking 2012:
Research only
User-friendly menu
More than 500 universities worldwide
Innovations:
* Top 10 % indicator of citation impact
* Collaboration indicators
* Fractional counting method (as default)
* Option to exclude non-English publications
* Stability intervals to represent uncertainty
28. www.leidenranking.com
Stability intervals clearly show the dramatic
influence of just one publication!
Examples: Göttingen, Utrecht
30. MNCS (earlier CPP/FCSm) is an average
and this is statistically not the best measure for a
skew distribution…
So we move on to p-top10%
this measures takes the whole distribution into
account
Notice: dimension of MNCS is C/P
dimension of p-top10% is n
33. Important observations:
* Large effect of all vs only-English publications
for D and F
* Large effect full vs fractional counting
* Remove one-paper-explosion effect by
MNCS > pTop10%
34. It appears that the medical fields benefit more
and the engineering fields less from the full
counting. Why? Answer will probably very
relevant for the future of rankings….
But also: this finding will be very influential in the
relative position of the engineering fields within a
university, particularly in performance analyses!
35. Largest Turkish universities in Leiden Ranking 2012
full all
Rank University Country P PPtop 10% PPtop 10% stability
1 Middle East Tech Univ 3766 8.7%
2 Ege Univ 4141 6.0%
3 Hacettepe Univ 6000 5.4%
4 Istanbul Univ 5744 4.8%
5 Gazi Univ 4647 4.5%
6 Ankara Univ 5096 4.2%
35
36. P MNCS
HACETTEPE UNIV ANKARA 6067 0,67
ISTANBUL UNIV 5819 0,65
ANKARA UNIV 5115 0,58
GAZI UNIV ANKARA 4671 0,64
EGE UNIV IZMIR 4172 0,76
MIDDLE EAST TECH UNIV ANKARA 3844 0,86
ISTANBUL TEKNIK UNIV 3013 0,92
DOKUZ EYLUL UNIV ISTANBUL 2851 0,71
ATATURK UNIV ERZURUM 2769 0,67
ONDOKUZ MAYIS UNIV 2521 0,50
ERCIYES UNIV 2509 0,81
SELCUK UNIV 2331 0,81
CUKUROVA UNIV ADANA 2175 0,68
FIRAT UNIV 2167 0,82
BASKENT UNIV 2150 0,50
For more than 40
MARMARA UNIV ISTANBUL 2085 0,71
ULUDAG UNIV 2032 0,61
GULHANE MIL MED ACAD 1884 0,53
Turkish universities KARADENIZ TECH UNIV 1884 0,71
SULEYMAN DEMUREL UNIV ISPARTA 1741 0,67
indicators are available AKDENIZ UNIV ANTALYA
BILKENT UNIV ANKARA
1722
1496
0,76
1,06
INONU UNIV 1402 0,66
BOGAZICI UNIV ISTANBUL 1360 0,82
DICLE UNIV 1255 0,56
PAMUKKALE UNIV 1215 0,73
KOCAELI UNIV 1211 0,77
TRAKYA UNIV 1167 0,47
YUZUNCU YILL UNIV 1139 0,49
YILDIZ TECH UNIV IST ANBUL 1114 0,87
UNIV GAZIANTEP 1098 0,61
MERSIN UNIV 1053 0,71
AFYON KOCATEPE UNIV 1046 0,56
KIRIKKALE UNIV 1004 0,71
ANADOLU UNIV 1000 0,99
CUMHURIYET UNIV 973 0,59
HARRAN UNIV 945 0,64
CELAL BAY AR UNIV 884 0,71
ADNAN MENDERES UNIV 860 0,53
CANAKKALE ONSEKIZ MART UNIV 839 0,46
MUSTAFA KEMAL UNIV ANTAKYA 829 0,56
ZONGULDAK KARAELMAS UNIV 819 0,57
GAZIOSMANPASA UNIV 809 0,94
36
KOC UNIV 747 1,15
ABANT IZZET BAY SAL UNIV 745 0,51
YEDITEPE UNIV 708 0,77
37. Trend: Turkish universities rapidly improve
their positions!
Leiden Ranking MNCS
2009 2010 2011
MNCS MNCS MNCS
HACETTEPE UNIV ANKARA 0,62 0,67 0,67
ISTANBUL UNIV 0,67 0,67 0,65
ANKARA UNIV 0,51 0,57 0,58
GAZI UNIV ANKARA 0,57 0,62 0,64
EGE UNIV IZMIR 0,64 0,76 0,76
MIDDLE EAST TECH UNIV ANKARA 0,64 0,80 0,86
ISTANBUL TEKNIK UNIV 0,81 0,90 0,92
DOKUZ EYLUL UNIV ISTANBUL 0,63 0,69 0,71
ATATURK UNIV ERZURUM 0,57 0,68 0,67
ERCIYES UNIV 0,57 0,77 0,81
ONDOKUZ MAYIS UNIV 0,41 0,48 0,50
SELCUK UNIV 0,61 0,75 0,81
FIRAT UNIV 0,64 0,74 0,82
CUKUROVA UNIV ADANA 0,56 0,65 0,68
BASKENT UNIV 0,44 0,53 0,50
37
38. Built on the basic data of the Leiden Ranking 2012:
Leiden Benchmark Analysis:
* all indicators of the ‘total’ university ranking are available for 30 main
(NOWT) fields of science
* their trends in the past 10 years
* world rank by indicator and by main field
All this in comparison with 20 other universities by choice
* all indicators for each field within each main field (total 200 fields)
with distinction between fields above and below university average
44. Conclusion:
Advanced bibliometric methods provide important,
valid, strategic information about the international
performance and the institutional structure of
universities and their departments
METU Ankara P[2007-2010], C[2007-2011]
48. Citation density map Neurology
citing-side normalization
Normalization based on ‘informal’ field:
references of the citing papers
No clinical research disadvantage because of
normalization with ‘own’ environment
48
52. Reassignment of multidisciplinary publications
• Publications in the Multidisciplinary sciences subject
category are reassigned to other subject categories
• Reassignment is done proportionally to the number
of references pointing to a subject category
• 9.4% of the multidisciplinary articles and reviews
from the period 2007–2011 cannot be reassigned
• Less than 0.2% of the articles and reviews in Nature
and Science cannot be reassigned
52
53. (1) selection of universities
Region:
* World
* Regions (Europe, Asia, North America, South
America, Africa, Oceania)
* Countries
By number of universities covered in the ranking:
*largest 100, 200, 300, 400, 500
54. (2) selection of performance
dimension with specific indicators
* impact: P, MCS, MNCS, PP(top10%)
* collaboration: P, PP(collab), PP(int collab),
MGCD, PP(long dist collab)
indicators with stability intervals!
55. (3) selection of calculation method
* full or fractional assignment collaborative
publications
* all WoS publications or only English language
71. Leiden University
Major field field P(2005-9) MNCS
P>100 CLINICAL MEDICINE MEDICINE, GENERAL & INTERNAL 291.2 3.03
rank by PHYSICS AND ASTRONOMY PHYSICS, MULTIDISCIPLINARY 192.5 2.65
MNCS PHYSICS AND ASTRONOMY PHYSICS, CONDENSED MATTER 204.0 2.10
CLINICAL MEDICINE RESPIRATORY SYSTEM 122.7 1.91
CLINICAL MEDICINE RHEUMATOLOGY 349.0 1.90
CLINICAL MEDICINE CARDIAC & CARDIOVASC SYSTEMS 488.5 1.84
BIOL SCI: HUMANS MEDICINE, RESEARCH & EXPER 100.0 1.81
CLINICAL MEDICINE SURGERY 207.9 1.73
CLINICAL MEDICINE UROLOGY & NEPHROLOGY 158.0 1.67
MOLECULAR BIOL & BIOCHEM BIOTECH & APPLIED MICROBIOL 106.8 1.66
SOCIAL SC RELATED TO MED PUBLIC, ENVIRONM & OCC HEALTH 111.6 1.65
PHYSICS AND ASTRONOMY ASTRONOMY & ASTROPHYSICS 814.5 1.65
BIOL SCI: HUMANS MICROBIOLOGY 179.3 1.58
CLINICAL MEDICINE GENETICS & HEREDITY 394.9 1.57
CLINICAL MEDICINE PERIPHERAL VASCULAR DISEASE 268.0 1.51
Leiden average MNCS = 1.45
Editor's Notes
Degree Correlations in an example of a more general phenomenon known as assortative mixing. Degree correlations have a strong effect on the structure of the network. Most social network have a postive correlations betwwen the degrees of adjancent nodes, non social ones have negative correlations. + core-periphery structure. The nodes with high degree are attracted with one another and so coagulate into a highly interconnected core surrounded by a periphery of lower-degree nodes. - High-degree nodes tend to be cattered more broadly over the network.