Upload
gary
View
41
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Scholarly network comparisons. Erjia Yan, Ying Ding, Cassidy Sugimoto. Backgrounds I. Motivation I. A higher level of research aggregate – the institution - is rarely studied An institution is a stable and representative unit to study the production, diffusion, and consumption of knowledge - PowerPoint PPT Presentation
Citation preview
Scholarly network comparisons
Erjia Yan, Ying Ding, Cassidy Sugimoto
Backgrounds I
Motivation I
• A higher level of research aggregate – the institution - is rarely studied
• An institution is a stable and representative unit to study the production, diffusion, and consumption of knowledge
• An institution is a distinct research entity which provides an opportunity for the combination of mappings from social, geographical, and cognitive perspectives.
Backgrounds II
Motivation II
• With the advancement of social network analysis, several types of scholarly networks are introduced to bibliometrics, such as citation networks, bibliographic coupling networks, cocitation networks, and coauthorship networks
• These networks have their own uses but currently we are unaware of the similarity among them
Dataset
• 59 journals indexed as the Information Science & Library Science category.
• All document types published within these journals from January 1965 to February 2010 were downloaded for analysis.
• Data were processed in two steps– To filter the dataset in order to create a local citation
network between institutions– To identify unique institution names from the affiliation
data
Network size
Time Size of institution citation networks
Size of paper citation networks
1991-2000 2,906*2,906 9,750*9,750
2001-2005 3,010*3,010 9,280*9,280
2006-2010 3,783*3,783 10,998*10,998
The construction of citation and coauthorship networks
The construction of cocitation and bibliographic coupling networks
The construction of topical networks
• Author-Conference-Topic (ACT) Model (Tang et al., 2008)
• Ten topics: • The topic similarity between two institutions can
be calculated through cosine similarity
• Sij is then the line value between institution i and institution j in the topical network
Clustering and mapping methods
• VOSviewer clustering and mapping (Waltman, Eck, & Noyons, 2010) technique is selected
• It is developed based on Clauset, Newman, and Moore’s (2004) algorithm for weighted networks.
Distance measurements
• Cosine distance (CD)
Distance measurements
• Earth mover’s distance (EMD)
Basic network characteristicsNo. oflinks
Sum of link weights
Density No. ofclusters
Size of the largest cluster
1991-2000
BGcoupling 84,676 277,199 0.0201 51 285Citation 31,280 49,652 0.0074 108 116Cocitation 70,618 210,583 0.0167 48 181Topic 460,809 460,809 0.1092 10 440Coauthor 5,260 6,232 0.0012 98 49
2001-2005
BGcoupling 100,498 399,688 0.0222 44 262Citation 40,073 65,102 0.0088 44 136Cocitation 89,596 259,871 0.0198 42 216Topic 647,980 647,980 0.1431 10 482Coauthor 7,127 8,969 0.0016 102 44
2006-2010
BGcoupling 248,934 873,446 0.0348 54 518Citation 64,750 103,556 0.0091 58 218Cocitation 134,951 474,673 0.0189 44 361Topic 686,196 686,196 0.0959 10 669Coauthor 11,729 14,609 0.0016 112 105
Clustering results of top institutionsIdx Institution name BGcoupling Citation Cocitation Topic Coauthor1 GEORGIA STATE UNIV,ATLANTA 4 29 4 6 202 HUNGARIAN ACAD SCI,HUNGARY 1 1 2 1 603 UNIV GEORGIA,ATHENS 9 42 4 6 884 UNIV MINNESOTA,MINNEAPOLIS 4 48 4 6 795 UNIV WESTERN ONTARIO,CANADA 15 6 1 9 856 INDIANA UNIV,BLOOMINGTON 7 15 33 9 947 FLORIDA STATE UNIV,TALLAHASSEE 7 30 28 9 478 UNIV BRITISH COLUMBIA,CANADA 5 16 4 6 1049 UNIV OKLAHOMA,NORMAN 20 11 30 2 13
10 UNIV SHEFFIELD,ENGLAND 2 13 26 3 1111 UNIV MARYLAND,COLLEGE PK 7 3 28 6 7512 UNIV MICHIGAN,ANN ARBOR 9 28 27 2 3713 DREXEL UNIV,PHILADELPHIA 14 42 13 9 3514 KATHOLIEKE UNIV LEUVEN,BELGIUM 1 1 2 1 6015 UNIV S FLORIDA,TAMPA 7 43 9 9 5716 ROYAL SCH LIB & INF SCI,DENMARK 1 9 9 3 7817 LEIDEN UNIV,NETHERLANDS 1 1 2 1 1418 UNIV ARIZONA,TUCSON 18 32 1 6 9719 UNIV PITTSBURGH,PITTSBURGH 12 5 3 6 7920 UNIV ILLINOIS,URBANA 2 6 1 2 70
bibliographic coupling network
citation network
cocitation network
coauthorship network
topical network
CD and EMD for each pair of networks
1991-2000 2001-2005 2006-2010CD EMD CD EMD CD EMD
BGcoupling-citation 0.74 0.34 0.72 0.39 0.77 0.37BGcoupling-cocitation 0.48 0.44 0.49 0.49 0.55 0.54BGcoupling-topic 0.90 0.30 0.89 0.33 0.90 0.32BGcoupling-coauthor 0.68 0.30 0.59 0.21 0.59 0.19Citation-cocitation 0.78 0.33 0.75 0.33 0.79 0.39Citation-topic 0.96 0.33 0.96 0.33 0.96 0.35Citation-coauthor 0.89 0.53 0.86 0.49 0.90 0.46Cocitation-topic 0.90 0.33 0.89 0.35 0.89 0.35Cocitation-coauthor 0.77 0.44 0.77 0.47 0.83 0.51Topic-coauthor 0.96 0.37 0.96 0.37 0.96 0.35
Ranking of network similaritiesBGcoupling Citation Cocitation Topic Coauthor
BGcoupling - 3 1 4 2Citation 1 - 2 4 3Cocitation 1 2 - 4 3Topic 2 3 1 - 4Coauthor 1 3 2 4 -
BGcoupling Citation Cocitation Topic CoauthorBGcoupling - 3 4 2 1Citation 2 - 3 1 4Cocitation 4 2 - 1 3Topic 1 2 3 - 4Coauthor 1 3 4 2 -
CD
EMD
Hybrid networks
• In order to capture both social and cognitive aspects of interactions of certain research aggregates, two types of networks, one from the social side and the other from the cognitive side, can be combined and thus forming a hybrid network.
• By considering the network density, we suggest the following combinations:– Coauthorship network and citation network;– Bibliographic coupling network and cocitation network; and – Bibliographic coupling network and topical network.