Dynamic research profile visualisation using cluster transition
Je n s Pe t e r An d e rs e n 1,2 Je s p e r W. Sc h n e id e r 1
1: Royal School of Library and Information Science, Aalborg, Denmark 2: Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science Medical Library Aalborg Hospital, Aarhus University Hospital
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
The setting
Aalborg Hospital is a regional hospital in the northern region of Denmark, and the latest danish hospital to achieve status of university hospital.
T h i sta tu s s ch a n g e i 2 0 0 3 n l d to a n e i cre a se d fo cu s n o n re se a rch a n d i n o va ti n . n o
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Aims
1.To describe the research and innovation profile and actual output of the various clinical and research departments 2.To investigate methods for illustrating changes over time
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
How we’ll get there
• The assessment framework • Materials & Methods • Results
– Research profiles – Clusters and transitions
• Implications
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Assessment framework
Publication indicators of: • Five composite score: Journal articles are awarded points based on journal (levels created based Funding [#PhD students, grants, research budgets] on field-normalised journal impact Research [publication score, dissertations] factor)
A: B: C: Mediation [media attendance, popular scientific articles]Points are fractioned and distributed among authors (department affiliations) D: Networking [conference participation, peer revieweing] E: Innovation [patents, consultancy, IPR-transactions] •
• [examples] • Departments are awarded scores in each category
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Methods - overview
• Profile creation (distribution of indicators) • Profile clustering – profile types • Visualisation of clusters over time • •
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Materials
• All clinical and research departments of Aalborg Hospital existing over the period of 2007-2009 included (34 departments) • • Data extracted from central databases. Data likely to be more complete (exact) in 2009 than 2007
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Methods – Department comparison
• Department scores can not be compared directly due to differences in e.g. department type (clinical vs. research) and size
•
• Distributions of scores, rather than absolute scores, allow for grouping => research profiles
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Methods – Profile clustering
• Profiles => (1x5) vectors • Aalborg Hospital: [Nx5] matrix per year • Vector comparison using cosine similarity => symmetric [NxN] similarity matrix • • Agglomerative, hierarchical clustering based on similarity matrix
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Methods – Profile clustering
Example of cluster dendrogram Similarities converted to distance (1-cosim) in order to use R agglomerative clustering method
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – Profile clusters
• Five clusters created in 2007 – naturally apparent threshold 0.8 – 1 outlier • Each cluster = profile type • • Clusters static over time, but approach splits in 2009
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – cluster transitions
2007
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 22 24 13 5 8 29 18 6 12 2 19 7 28 4 34 27 32 10 31 3 20 26 11 17 30 25 9 16 1 15 33 14 21
2008
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 17 9 34 10 27 22 4 24 8 29 5 32 28 30 16 18 12 6 19 31 2 7 3 33 21 20 13 14 11
2009
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 29 5 4 8 31 18 2 16 34 9 28 30 7 12 19 33 32 6 24 17 22 14 27 10 3 1 25 11
A
A
B
B
C
E
DEP 26 DEP 25 DEP 23 DEP 1 DEP 15
D
DEP DEP DEP DEP
21 20 13 15
E
E
D
DEP 23
C
DEP 26 DEP 23
D
C
B
A
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – cluster transitions
2007
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 22 24 13 5 8 29 18 6 12 2 19 7 28 4 34 27 32 10 31 3 20 26 11 17 30 25 9 16 1 15 33 14 21 17 Example of profile,DEPtype ”E”: DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 9 34 10 27 22 4 24 8 29 5 32 28 30 16 18 12 6 19 31 2 7 3 33 21 20 13 14 11
2008
2009
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 29 5 4 8 18 2 16 34 9 28 30 7 12 19 33 32 6 24 17 22 14 27 10 3 1 25 11
A
Top-research output, low focus on other areas 31 DEP
A
B
B
C
E
DEP 26 DEP 25 DEP 23 DEP 1 DEP 15
D
DEP DEP DEP DEP
21 20 13 15
E
E
D
DEP 23
C
DEP 26 DEP 23
D
C
B
A
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – cluster transitions
2007
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 22 24 13 5 8 29 18 6 12 2 19 7 28 4 34 27 32 10 31 3 20 26 11 17 30 25 9 16 1 15 33 14 21
2008
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 17 9 34 10 27 22 4 24 8 29 5 32 28 30 16 18 12 6 19 31 2 7 3 33 21 20 13 14 11
2009
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 29 5 4 8 31 18 2 16 34 9 28 30 7 12 19 33 32 6 24 17 22 14 27 10 3 1 25 11
A
A
Example of profile, type ”A”: High networking, medium research
B
B
C
E
DEP 26 DEP 25 DEP 23 DEP 1 DEP 15
D
DEP DEP DEP DEP
21 20 13 15
E
E
D
DEP 23
C
DEP 26 DEP 23
D
C
B
A
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – cluster transitions
2007
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 22 24 13 5 8 29 18 6 12 2 19 7 28 4 34 27 32 10 31 3 20 26 11 17 30 25 9 16 1 15 33 14 21
2008
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 17 9 34 10 27 22 4 24 8 29 5 32 28 30 16 18 12 6 19 31 2 7 3 33 21 20 13 14 11
2009
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 29 5 4 8 31 18 2 16 34 9 28 30 7 12 19 33 32 6 24 17 22 14 27 10 3 1 25 11
A
A
B
B
C
E
DEP 26 DEP 25 DEP 23 DEP 1 DEP 15
D
DEP DEP DEP DEP
21 20 13 15
E
E
D
DEP 23
C
DEP 26 DEP 23
D
C
B
A
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Results – cluster transitions
2007
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 22 24 13 5 8 29 18 6 12 2 19 7 28 4 34 27 32 10 31 3 20 26 11 17 30 25 9 16 1 15 33 14 21
2008
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 17 9 34 10 27 22 4 24 8 29 5 32 28 30 16 18 12 6 19 31 2 7 3 33 21 20 13 14 11
2009
DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP DEP 29 5 4 8 31 18 2 16 34 9 28 30 7 12 19 33 32 6 24 17 22 14 27 10 3 1 25 11
A
A
B
B
C
E
DEP 26 DEP 25 DEP 23 DEP 1 DEP 15
D
DEP DEP DEP DEP
21 20 13 15
E
E
D
DEP 23
C
DEP 26 DEP 23
D
C
B
A
Det Informationsvidenskabelige Akademi Royal School of Library and Information Science
Implications & conclusions
• Detection of potential problems – but not causes • Visualisation of several indicators collected in few, simple diagrams • Additional intellectual analysis necessary •
Jens Peter Andersen Royal School of Library and Information Science email: jpa@iva.dk http://www.iva.dk
The End
THANK YOU FOR YOUR ATTENTION
