=Paper=
{{Paper
|id=None
|storemode=property
|title=The Collaboration Potential, an Index to Assess the Roles of Scientists in their Coauthorship Networks
|pdfUrl=https://ceur-ws.org/Vol-835/paper23.pdf
|volume=Vol-835
|dblpUrl=https://dblp.org/rec/conf/iir/GiulianiPN12
}}
==The Collaboration Potential, an Index to Assess the Roles of Scientists in their Coauthorship Networks==
https://ceur-ws.org/Vol-835/paper23.pdf
The Collaboration Potential, an index to assess
the roles of scientists in their coauthorship
networks
Francesco Giuliani1 , Michele Pio De Petris1 and Giovanni Nico2
1
Innovation and Technological Development, IRCCS Casa Sollievo della Sofferenza,
San Giovanni Rotondo, Italy,
f.giuliani@operapadrepio.it, m.depetris@operapadrepio.it,
2
Istituto per le Applicazioni del Calcolo, Consiglio Nazionale delle Ricerche
(CNR-IAC), Italy, g.nico@ba.iac.cnr.it
Abstract. Over the past decade there have been many investigations
aimed at defining the role of scientists in their coauthorship networks.
In this work we propose an analytical definition of a collaboration po-
tential between authors of scientific papers based on both coauthorships
and content sharing. The collaboration potential can also be considered
a tool to investigate the weakness of the network in terms of ‘lost collab-
orations’ between authors with same scientific interests. This work is an
abbreviated version of the original article from the same authors [1].
Keywords: social network analysis, scientific collaborations, coauthor-
ships, collaboration potential
1 Introduction and methodological approach
In this work we present a method aimed at investigating the informative potential
that modern bibliographic databases offer. We study the publication output of
researchers and try to find an index describing both collaborations and content
sharing in scientific networks.
Considering coauthorship as synonymous of collaboration, we can define a
collaboration index between author A and author B as
dim(PA ∩ PB )
PAB = , (1)
dim(PA )
where PA and PB represent the sets of papers authored by A and B respectively,
dim(PA ) represents the number of elements (articles) authored by author A and
dim(PA ∩ PB ) represents the number of articles shared by authors A and B as
coauthors. This index represents for author A the fraction of articles he has
written in collaboration with author B. This index, taken alone, does not tell
the whole story about collaboration as it is independent from article contents.
One can build an index to express content sharing defining it as the number
of keywords author A and author B share divided by the number of keywords of
author A. This index is a measure of the commonality of scientific interests, but
does not take into account collaborations between authors. If we want to measure
the collaboration potential between two authors we need to build a consistent
index taking into account both coauthored papers and contents of such papers,
that in our model are represented by article keywords.
We want keywords to come from an unambiguous and limited set of terms,
so we chose to study only publications indexed by the PubMed search en-
gine. For such publications, keywords come from MeSH (Medical Subject Head-
ings) database, a controlled vocabulary thesaurus used for indexing articles in
PubMed. We simply used a custom query and XML parsing in order to associate
keywords to articles of our interest.
2 Measuring the collaboration potential
In our simple model we start taking coauthorships into account. We can define
the set of articles author A has not co-authored with author B (and vice versa)
as
P A = PA − (PA ∩ PB ) , P B = PB − (PA ∩ PB ) (2)
The articles belonging to these sets are associated with their respective keywords,
i.e. we can define the sets K A and K B containing the keywords of the papers the
two authors have not respectively coauthored. The intersection between these
two sets
K AB = K A ∩ K B , (3)
represents the keywords shared by the articles the authors have not co-authored.
So we can formulate the collaboration potential based on non-coauthorship for
author A towards author B as
dim(K A ∩ K B )
mAB = . (4)
dim(K A )
This index has many interesting characteristics. It is defined in the interval [0, 1]
and is 0 in two circumstances:
– First case: the two authors have coauthored all their articles. In this case
the sets K A and K B are both void so dim(K A ∩ K B ) is 0, the authors
having fully exploited their collaboration potential, having co-authored all
they could, i. e. all the articles they wrote.
– Second case: for the articles they have not coauthored, they worked on totally
different subjects. In this case K A ∩ K B is void meaning that the authors,
excluding coauthored articles, share no common scientific interests and so,
according to our model, no collaboration potential exists between them.
We can discriminate between the two cases in which mAB = 0 according to
the corresponding value of PAB . In fact a value PAB = 1 corresponds to the first
case, while a PAB = 1 to the second case.
In all other cases mAB different from 0 implies the existence of a not fully
“exploited” collaboration between authors A and B. The other extreme value of
the index is 1. In this case K A = K AB = K B , i.e. author A and author B share
all their keywords for the articles they have not coauthored. It is worth noting
that the collaboration potential we’ve just defined should not be considered a
“predictor” of future collaborations but it is intended to investigate the role of
scientists in the collaboration network. Other approaches were proposed in the
literature [2], [3] and methods were presented in order to predict the evolution
of links in a social network based on topology taken alone. Our method is quite
different because it relies on intrinsic node properties (identified as keywords),
and tries to investigate properties of links in terms of ‘lost collaboration’ between
the authors.
Extending 4 we can easily compute the collaboration potential between au-
thor A and group G considering the group as a single author, i. e. considering
the set of articles written by author A and the set of articles written by all other
authors of group G. We thus obtain:
dim(K A ∩ K G )
mAG = , (5)
dim(K A )
where K G represents the set of keywords for the articles author A has not coau-
thored with the other authors belonging to group G. If author A belongs to
group G the index in (5) expresses the collaboration potential the author has
with the colleagues of his own group, supposedly studying the same subjects of
his researches and publications.
3 Application of the method and discussion of the results
To apply our method we considered the publications and authors of the Casa
Sollievo della Sofferenza research hospital in years 2004-2009. For all authors
(216) and publications (711 papers, with a mean of 14.42 keywords per article)
we computed the collaboration potential according to eqs (4) and (5). We found
a mean value for PAG of 90.50%, confirming that scientists coauthor the largest
majority of their publications with authors of their own group than with authors
belonging to other research groups of the institute.
In order to investigate the role of scientists inside and between research
groups, we considered the values of PAG and of mAG for each researcher (see
fig.1). The majority of authors concentrate on the bottom-right area of the plot.
This result confirms that generally authors have a low collaboration potential
with colleagues of their groups. The value of the collaboration potential is ex-
actly zero for 81.48% of authors. This result is simply understandable in terms
of coauthorships, in fact we have found that in all cases in which mAG is zero
PAG is one, meaning that each of these authors’ publication is coauthored by at
least one other author of the group the author belongs to. We could define these
authors as highly integrated with their research units, writing their papers with
at least one of the colleagues of their groups. Furthermore, we found a small
subset of authors having a low value of PAG and an high value of mAG with
their group. We can easily define these authors as “independent” as they share
no article with the members of their own group, given many subjects on which
they “could” have written articles together.
Eventually, generalizing the concept of collaboration to a broader scope, the
methods presented herein could easily be used to define a collaboration potential
in every case in which one can classify the content of some activities and deter-
mine which of them are in common among the actors cooperating to perform
such activities.
Fig. 1. Distribution of authors according to the collaboration potentials toward their
research groups (mAG ) and coauthorship sharing (PAG ) values. The grey bar graphs
on the axes show the frequency distributions of the number of authors for each interval
of (mAG ) and (PAG ).
References
1. Giuliani F, De Petris MP, Nico G, Assessing scientific collaboration through coau-
thorship and content sharing, Scientometrics, 85(1), p13–28, October 2010.
2. David Liben-Nowell and Jon Kleinberg, The Link-Prediction Problem for Social
Networks, Journal of the American Society for Information Science and Technology,
58(7), p1019–1031, May 2007.
3. Leo Katz, A new status index derived from sociometric analysis, Psychometrika,
18(1), p39–43, March 1953.