Title: An Experimental Comparison of Bibliometric Mapping Techniques
1An Experimental Comparison of Bibliometric
Mapping Techniques
- Nees Jan van Eck, Ludo Waltman, Rommert Dekker
- Erasmus University Rotterdam, The Netherlands
- nvaneck,lwaltman,rdekker_at_few.eur.nl
- Jan van den Berg
- Delft University of Technology, The Netherlands
- j.vandenberg_at_tudelft.nl
- 10th International Conference on Science and
Technology Indicators - Vienna, September 18, 2008
1
1
1
1
2Bibliometric mapping
Unit of analysis Unit of analysis Unit of analysis Unit of analysis Unit of analysis
Authors Journals Words/terms Web pages
Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure
Direct Direct Direct Direct Indirect Indirect Indirect
Jaccard Cosine Association strength Pearson correlation Cosine
Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique
Distance based Distance based Distance based Distance based Graph based Graph based Graph based
MDS VxOrd VOS Pajek Pathfinder networks
3Bibliometric mapping
Unit of analysis Unit of analysis Unit of analysis Unit of analysis Unit of analysis
Authors Journals Words/terms Web pages
Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure
Direct Direct Direct Direct Indirect Indirect Indirect
Jaccard Cosine Association strength Pearson correlation Cosine
Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique
Distance based Distance based Distance based Distance based Graph based Graph based Graph based
MDS VxOrd VOS Pajek Pathfinder networks
4Data sets
Unit of analysis Unit of analysis Unit of analysis Unit of analysis Unit of analysis
Authors Journals Words/terms Web pages
- Journals in economics, management, and operations
research - 2001-2005
- ISI categories business business, finance
economics management and operations research
management science - 376 journals
- Terms in computational intelligence
- 2001-2005
- 337 terms
Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure
Direct Direct Direct Direct Indirect Indirect Indirect
Jaccard Cosine Association strength Pearson correlation Cosine
Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique
Distance based Distance based Distance based Distance based Graph based Graph based Graph based
MDS VxOrd VOS Pajek Pathfinder networks
5Similarity measures
Unit of analysis Unit of analysis Unit of analysis Unit of analysis Unit of analysis
Authors Journals Words/terms Web pages
Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure
Direct Direct Direct Direct Indirect Indirect Indirect
Jaccard Cosine Association strength Pearson correlation Cosine
- Other names for the association strength
- probabilistic affinity index
- proximity index
Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique
Distance based Distance based Distance based Distance based Graph based Graph based Graph based
MDS VxOrd VOS Pajek Pathfinder networks
6Mapping techniques
Unit of analysis Unit of analysis Unit of analysis Unit of analysis Unit of analysis
Authors Journals Words/terms Web pages
- Ordinal MDS in SPSS PROXSCAL
- VxOrd (DrL)
- VxOrd1 top 10 similarities, edge cutting
(default settings) - VxOrd2 all similarities, no edge cutting
- VxOrd3 top 10 similarities, no edge cutting
Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure Similarity measure
Direct Direct Direct Direct Indirect Indirect Indirect
Jaccard Cosine Association strength Pearson correlation Cosine
Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique Mapping technique
Distance based Distance based Distance based Distance based Graph based Graph based Graph based
MDS VxOrd VOS Pajek Pathfinder networks
7Evaluation criteria
Quality of the maps
Does a map provide a satisfactory representation of the data?
Robustness
Measures the similarity between maps that have been constructed based on different samples drawn randomly from the same data set
Efficiency
Measures the similarity between a map that has been constructed based on an entire data set and maps that have been constructed based on samples drawn randomly from that data set
8MDS Jaccard index (journals)
9VxOrd1 cosine (journals)
10VxOrd2 cosine (journals)
11VxOrd3 cosine (journals)
12VOS association strength (journals)
13Comparison of maps (journals)
- MDS clearly performs worst, with no good
separation between journals from different fields - As is typical for MDS, journals are located more
or less uniformly distributed within a circle - VxOrd3 and VOS perform quite well, with a clear
clustering of journals based on their field
13
14Robustness and efficiency (journals)
- MDS and VxOrd1 have a low robustness and
efficiency - The other techniques do much better
14
15MDS Jaccard index (terms)
16VxOrd1 cosine (terms)
17VxOrd2 cosine (terms)
18VxOrd3 cosine (terms)
19VOS association strength (terms)
20Comparison of maps (terms)
20
21Robustness and efficiency (terms)
21
22Conclusions
MDS VxOrd VOS
Quality of the maps locates objects more or less uniformly distributed within a circle, with no clearly separated clusters produces attractive maps with a very clear clustering, but only if appropriate parameter values are chosen produces satisfactory maps without the need to choose any parameter values
- No complete specification of VxOrd is available
in the literature
Robustness and efficiency low depends on parameter values and data set high
22