<abstract xmlns="http://www.w3.org/1999/xhtml"><sec id="j_jdis.201620_s_006_w2aab2b8c62b1b7b1aab1c15b1Aa"><h3>Purpose</h3><p> This study aims to answer the question to what extent different types of networks can be used to predict future co-authorship among authors.</p></sec><sec id="j_jdis.201620_s_007_w2aab2b8c62b1b7b1aab1c15b2Aa"><h3>Design/methodology/approach</h3><p>We compare three types of networks: unweighted networks, in which a link represents a past collaboration; weighted networks, in which links are weighted by the number of joint publications; and bipartite author-publication networks. The analysis investigates their relation to positive stability, as well as their potential in predicting links in future versions of the co-authorship network. Several hypotheses are tested.</p></sec><sec id="j_jdis.201620_s_008_w2aab2b8c62b1b7b1aab1c15b3Aa"><h3>Findings</h3><p> Among other results, we find that weighted networks do not automatically lead to better predictions. Bipartite networks, however, outperform unweighted networks in almost all cases.</p></sec><sec id="j_jdis.201620_s_009_w2aab2b8c62b1b7b1aab1c15b4Aa"><h3>Research limitations</h3><p> Only two relatively small case studies are considered.</p></sec><sec id="j_jdis.201620_s_010_w2aab2b8c62b1b7b1aab1c15b5Aa"><h3>Practical implications</h3><p> The study suggests that future link prediction studies on co-occurrence networks should consider using the bipartite network as a training network.</p></sec><sec id="j_jdis.201620_s_011_w2aab2b8c62b1b7b1aab1c15b6Aa"><h3>Originality/value</h3><p> This is the first systematic comparison of unweighted, weighted, and bipartite training networks in link prediction.</p></sec></abstract>

Purpose This study aims to answer the question to what extent different types of networks can be used to predict future co-authorship among authors.Design/methodology/approachWe compare three types of networks: unweighted networks, in which a link represents a past collaboration; weighted networks, in which links are weighted by the number of joint publications; and bipartite author-publication networks. The analysis investigates their relation to positive stability, as well as their potential in predicting links in future versions of the co-authorship network. Several hypotheses are tested.Findings Among other results, we find that weighted networks do not automatically lead to better predictions. Bipartite networks, however, outperform unweighted networks in almost all cases.Research limitations Only two relatively small case studies are considered.Practical implications The study suggests that future link prediction studies on co-occurrence networks should consider using the bipartite network as a training network.Originality/value This is the first systematic comparison of unweighted, weighted, and bipartite training networks in link prediction.

Predictive Characteristics of Co-authorship Networks: Comparing the Unweighted, Weighted, and Bipartite Cases

Journal of Data and Information Science

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Purpose This study aims to answer the question to what extent different types of networks can be used to predict future co-authorship among...

		UA	INF
Training	Number of authors	1,102	397
	Number of papers	7,569	1,118
	Average number of papers per author	11.9	4.3
	Average number of authors per paper	1.7	1.5
Test	Number of authors	1,102	397
	Number of papers	7,939	1,069
	Average number of papers per author	12.8	4.4
	Average number of authors per paper	1.8	1.6

	UA		INF

	F_max	P@20	F_max	P@20
Reoccurrence	0.59	0.48	0.61	0.48
Reoccurrence with weights	0.59	1.00	0.61	0.81
Reoccurrence with author numbers	0.55	0.52	0.61	0.57

		Unweighted	Weighted	Bipartite
Common neighbors	F_max	0.3964	0.3928	0.5866
	P@20	1	1	1
Cosine	F_max	0.4025	0.397	0.5865
	P@20	0.1429	0.0952	0.7619

		Unweighted	Weighted	Bipartite
Common neighbors	F_max	0.4091	0.3935	0.6109
	P@20	0.8571	0.5714	0.8095
Cosine	F_max	0.4332	0.4305	0.6217
	P@20	0.0476	0.0476	1

Predictive Characteristics of Co-authorship Networks: Comparing the Unweighted, Weighted, and Bipartite Cases

Article Category: Research Paper

Published Online: Sep 01, 2017

Page range: 59 - 78

Received: May 17, 2016

Accepted: Jun 26, 2016

DOI: https://doi.org/10.20309/jdis.201620

Keywords
Network evolution, Link prediction, Weighted networks, Bipartite networks, Two-mode networks

© 2016 Raf Guns

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Descriptive statistics of the two datasets.

Results for positive stability.

Results for neighborhood-based predictors (UA).

Results for neighborhood-based predictors (INF).

Predictive Characteristics of Co-authorship Networks: Comparing the Unweighted, Weighted, and Bipartite Cases

Article Category: Research Paper

Published Online: Sep 01, 2017

Page range: 59 - 78

Received: May 17, 2016

Accepted: Jun 26, 2016

DOI: https://doi.org/10.20309/jdis.201620

KeywordsNetwork evolution, Link prediction, Weighted networks, Bipartite networks, Two-mode networks

© 2016 Raf Guns

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Descriptive statistics of the two datasets.

Results for positive stability.

Results for neighborhood-based predictors (UA).

Results for neighborhood-based predictors (INF).

Keywords
Network evolution, Link prediction, Weighted networks, Bipartite networks, Two-mode networks