A Geographical Analysis of Knowledge Production in Computer Science Guilherme Vale Menezes Nivio Ziviani Alberto H. F. Laender Virgílio Almeida [email protected].

A Geographical Analysis of Knowledge Production in Computer Science

Guilherme Vale Menezes

Nivio Ziviani

Alberto H. F. Laender

Virgílio Almeida

[email protected]

[email protected]

[email protected]

[email protected]

Federal University of Minas Gerais - Brazil

LAboratory for Treating INformation (LATIN) – UFMG - Brazil

Summary

Introduction Data Gathering Results Conclusions

The Problem

Study the characteristics of researchers of Computer Science graduate programs

30 graduate programs in 3 geographic regions Build collaboration social networks based on

DBLP We use several metrics of collaboration social

networks Giant Component Clustering Coefficient


Steps

Comparison between 30 programs in 3 regions Comparison between 30 Computer Science fields Study of the interrelationship between fields Temporal analysis of the 3 regions and the fields


Collaboration Network

Author

Collaboration


Collaborations in DCC-UFMG


Collaborations in DCC-UFMG


Summary



Data Gathering

Part of our data came from Perfil-CC project Objective of Perfil-CC: study Brazilian Computer

Science graduate programs A set of 30 programs was chosen Focus: comparison with North American programs Results supported public policies Data gathered in June 2007


Brazilian Programs

8 graduate programs391 authors

PUC-Rio, UFRJ, UFPE, UFMG,USP-SP, USP-SC, UNICAMP, UFRGS


Canadian and US Programs

16 graduate programs1,262 authors

British Columbia, Toronto, Waterloo, Brown, CalTech,CMU, Cornell, Harvard, Illinois, MIT, Princeton, Stanford, UC Berkeley, UTexas Austin, Washington, Wisconsin


French, Swiss and UK Programs

6 graduate programs611 authors

ETH Zurich, Cambridge U., Imperial College, Oxford U., École Polytechnique, Paris VI


Data Gathering Professors obtained from the departments’ websites

Publications from DBLP

Programs: 30Professors: 2,007Authors: 76,537Papers: 352,766Venues: 2,176


Venue Classification

2,176 were classified (by humans) into 30 fields The list of fields was obtained from a poll The brazilian Computer Science research

community was consulted 312 researchers identified 30 different fields


Computer Science Fields

Algorithms and Theory Applied Computing Artificial Intelligence Bioinformatics

Circuit Conception Comp. Architecture Comp. Graphics Comp. Networks and Distributed Systems

Computer Vision Data Mining Databases Embedded and Real-Time Systems

Formalisms, Logics and Semantics

Games and Entertainment

Geoinformatics Human-Comp. Interaction

Informatics in Education

Information Retrieval Information Systems Machine Learning

Modelling and Simulation

Natural Language Processing

Operation Systems Operational Research and Optimization

Programming Languages

Robotics, Automation and Control

Security and Privacy Software Engineering

Ubiquitous Comp. Web, Hypermedia, Multimedia



















Algorithms and Theory


















InformationRetrieval


















Bioinformatics

Summary



General Statistics

Br Ca-US Fr-Sw-UK

Authors 357 1,008 488


General Statistics

Br Ca-US Fr-Sw-UK

Authors 357 1,008 488

Papers 4,405 40,039 8,764


General Statistics

Br Ca-US Fr-Sw-UK

Authors 357 1,008 488

Papers 4,405 40,039 8,764

Paper/Author 16.06 45.89 19.85


General Statistics

Br Ca-US Fr-Sw-UK

Authors 357 1,008 488

Papers 4,405 40,039 8,764

Paper/Author 16.06 45.89 19.85

Author/Paper 3.21 2.87 2.77


Giant Component A connected component is a maximum connected

subgraph


Giant Component


A connected component is a maximum connected subgraph

Giant Component

Giant Component size = 5 / 11 = 0.45 = 45%


A connected component is a maximum connected subgraph

The largest connected component is the giant component

Giant Component

Br Ca-US Fr-Sw-UK

0.7815 0.7827 0.2617


Giant Component inside ProgramsBr Ca-US Fr-Sw-UK

PUC-Rio 0,83 Wisconsin 0,71 Oxford 0,46UFMG 0,76 Harvard 0,70 Imperial College 0,40

USP-SC 0,73 Cornell 0,66École

Polytechnique0,30

UFRGS 0,70 Stanford 0,66 ETH Zurich 0,26UFPE 0,68 Washington 0,65 Cambridge 0,25UFRJ 0,64 Illinois 0,63 Paris IV 0,12

UNICAMP 0,62 Princeton 0,63 - -USP-SP 0,43 B. Columbia 0,62 - -

- - U Texas 0,62 - -- - CMU 0,59 - -- - UC Berkeley 0,59 - -- - Brown 0,53 - -- - MIT 0,49 - -- - Waterloo 0,42 - -- - Caltech 0,21 - -

Average 0,67 Average 0,56 Average 0,30


Clustering Coefficient



Clustering coefficient of the network is the average clustering coefficient of its vertexes

The clustering coefficient is a measure of transitivity



Br Ca-US Fr-Sw-UK

0.30 0.20 0.38



Clustering Coefficient below the average (87%) for fields closely related to Mathematics Algorithms and Theory (79%) Operational Reaseach and Optimization (83%) Formalisms, Logics and Semantics (83%)


Interrelationship between Fields


Interrelationship between Fields


Giant Component Evolution






Increase in the number of graduate programs in 1990s



Increase in government funding



A shift in policy: more support to research groups instead of individuals







Edges vs Vertices


Br Ca-US Fr-Sw-UK

Inclination 1.39 1.84 1.23

2 well-established fields Computer Architecture Databases

2 emerging fields Bioinformatics Geoinformatics

Clustering Coefficient Evolution






Densification

Summary



Conclusions Analysis of the characteristics of researchers of

Computer Science graduate programs


Analysis of the characteristics of researchers of Computer Science graduate programs

Differences in the collaboration network of Br, Ca-US and Fr-Sw-UK Giant component Clustering coefficient

Conclusions


Analysis of the characteristics of researchers of Computer Science graduate programs

Differences in the collaboration network of Br, Ca-US and Fr-Sw-UK Giant component Clustering coefficient

Smaller clustering coefficient for areas more closely related to Mathematics

Conclusions


Conclusions

Fast growth of the giant component in Brazil


Conclusions


The number of edges grows faster than the number of vertices in the three regions; faster growth in Ca-US


Conclusions


The number of edges grows faster than the number of vertices in the three regions; faster growth in Ca-US

Densification of emerging fields


ReferencesLaender, Lucena, Maldonado, Souza e Silva, Ziviani. Assessing the Research and Education Quality of the Top Brazilian Graduate Programs. ACM SIGCSE Bulletin, 40:135-145, June 2008.

Martins, Gonçalves, Laender, Ziviani. Assessing the Quality of Scientific Conferences Based on Bibliographic Citations. Scientometrics, to appear. 2009.


Guilherme Vale Menezes

Nivio Ziviani

Alberto H. F. Laender

Virgílio Almeida

[email protected]

[email protected]

[email protected]

[email protected]

Federal University of Minas Gerais - Brazil

?

A Geographical Analysis of Knowledge Production in Computer Science Guilherme Vale Menezes Nivio Ziviani Alberto H. F. Laender Virgílio Almeida [email protected].

Documents

dccufmg laboratory

theory slide

multimedia laboratory

different fields laboratory

brazilian programs

wisconsin laboratory

ufrgs laboratory

computer science fields