Development of scientific theories and technology is a result of continuous interaction, creation, and effective diffusion of ideas between researchers in the knowledge ecosystem. Digitization of publications and advancements in communication technology have made it easier for researchers to be aware of the existing knowledge capital and possible gaps in the field of study. This facilitation of the spread of scientific knowledge helps researchers to refine their own research methods and to contextualize their work within the domain. It also establishes an indirect interaction between individuals. One might not know a researcher personally but is still aware of her work through technical literature and can gain a sense of familiarity with it. Researchers attend gatherings and conferences to broaden their scope of a subject area and look for new ideas and open problems. Awareness to the state of the art and motivation to solve open problems often becomes a factor in setting up new collaborations between individuals.
Publication of scientific articles provides a narrow, but a well-quantified record of collaboration and exchange of technical information. Interactions between researchers can either be by citing one another or by co-authoring papers together, resulting in a complex system that is changing over time. With such a complex and dynamic system at hand, it is interesting to look for any possible underlying pattern hidden in interactions between researchers, and if these interactions have any mathematical structure. It is possible that such a structure can be used to explain the spread of knowledge and the growth of research fields.
Tools and methods developed within the framework of network science have proven to be very effective in addressing questions of such nature both quantitatively and qualitatively Newman (2001a, b); Sinatra et al. (2016). Effective treatment of complex systems by using networks and easy access to huge databases of publications have attracted a lot of attention in the last two decades. It has also lead to enormous research on structure and evolution of scientific collaboration Dong et al. (2017); Tomassini and Luthi (2007).
Detailed publication records make it easier to create citation networks and co-authorship networks. While the former are directed networks where edges represent citations between nodes, the latter are undirected (either weighted or unweighted) networks where edges exist between two nodes sharing an authorship on a paper. Nodes in these networks can either be papers, authors, universities, and so on, depending on the research question of interest.
Individual relationships between authors have a great impact on interactions between institutions at meso-level, and countries at macro-level Tomassini and Luthi (2007), shaping the changing trend of collaboration. Over time, the pattern of collaboration has shown a shift from individual efforts to more cooperative research, increasing the productivity and diversity of scientific publications globally, and resulting in an increase of innovation in this century Dong et al. (2017). Co-authorship networks have been shown to exhibit small-world characteristics and display high levels of clustering Newman (2004). As the network grows and new authors appear, the network structure changes Huang et al. (2008); Martin et al. (2013); Vasques Filho (2018). Growing networks also change authors topological position in the network structure, which is found to be directly related to one’s productivity and popularity Newman (2001a, b, 2004). The mechanism for evolving co-authorship networks has shown to exhibit an underlying preferential attachment process Barabási et al. (2002); Chen et al. (2013). It is observed that the co-authorship network becomes more connected over time, indicating an increase in collaboration between authors. As a result, the average node distance in the network shrinks. Knowing the co-authorship network structure and its evolution also makes it possible to predict future links between authors by exploiting the changes in an author’s neighborhood structure Huang et al. (2008). Studying co-authorship networks can explain the emergence of new research groups, the significance of some lead researchers, and how one’s collaborators change over time.
Citation networks, on the other hand, indicate the patterns in generation and diffusion of ideas in the scientific community. Citations received by publications play a significant role in determining their impact as well as their authors’ significance in the community Sinatra et al. (2016). Evolution of citing patterns can be correlated with the evolution of research fields Gualdi et al. (2011); Shi et al. (2009). Since citation networks are directed, tree-like hierarchical structures form the backbone of citations. Patterns in citation dynamics have been extensively explored and modeled. The key feature in citation patterns is the presence of a delay before a paper receives initial citations. Citations acquired by a paper typically increase shortly after publication and reach a maximum within the first few years before decaying with time Pan et al. (2018); Higham et al. (2017). Considering that the aging effect is important to quantify the probability of getting cited Börner et al. (2004) or the strength of collaboration in citations and co-authorship networks respectively. The effect of aging can be accounted for in different ways. It can be a weighted measure on edges, that decays since the contributing authors last shared publication Fiala et al. (2015) proportional to the time difference between simultaneous participation Tutoky and Paralič (2011). Instead of decaying weights on the edges, adding an aging effect on nodes in the citation networks can also determine the changing probability of receiving citation Hajra and Sen (2005).
Co-authorship and citation networks together reflect the structure and growth of scientific collaboration. Every new publication results in co-authorship and citation events, therefore it is intuitive that both citation and co-authorship networks complement each other and should have a strong positive correlation in their respective evolution. Many studies addressed these networks and pointed out strong interdependent relations between evolving citations and associated co-authorship networks Kas et al. (2012); Keegan et al. (2013); Martin et al. (2013); Amblard et al. (2011); Ding (2011); Tol (2011); Glänzel and Thijs (2004). Network measures on time varying graphs for both citation and co-authorship networks exhibit co-dependence of these networks in citing patterns and formation of communities Amblard et al. (2011). Topic modelling algorithms used to assign topics to papers and to investigate citation patterns between authors from similar and different topics fields showed close collaboration between authors working on similar topics. Also, that high profile authors do not generally co-author with one another but do closely cite each other Ding (2011). Large scale data set studies Wallace et al. (2012); Martin et al. (2013) solidify the notion of strong interdependence between citation and co-authorship networks. Citation exchange calculated up to a limited depth of co-authorship connections reveal large gaps in citing patterns of co-authors between natural sciences and social sciences and that the rate of self-citations is constant Wallace et al. (2012). A detailed analysis of citation and co-authorship networks constructed from a large longitudinal data set (100 years) of publications in Physical Review journals investigated the temporal changes in citing patterns between collaborators Martin et al. (2013). One of the main findings of the latter was the constant nature of the fraction of self-citations and citations among co-authors with a strong tendency towards reciprocal citations.
The existing interdependence between the two networks has also helped to define sophisticated weighted measures to distribute the credit of citations between co-authors of a paper, resulting in a more efficient way to calculate authors’ significance Tol (2011). Studying the citation and co-authorship networks simultaneously not only helps in quantifying a researcher’s contribution to the field González-Teruel et al. (2015) — network centrality measures have also proved to be important in quantifying the effect of citation and co-authorship networks on each other Biscaro and Giupponi (2014). It has been observed that an author’s (node’s) centrality value in the co-authorship network is a significant factor behind the number of citations received by them Sarigöl et al. (2014). Considering the effects of co-authorship networks is also important to define more sophisticated growth mechanisms for citing patterns in networks Guo et al. (2017). Combined co-authorship and citation networks have been used to predict new collaboration opportunities; that is, new edges in co-authorship networks Lande and Andrushchenko (2016) and also to quantify effect of authors and their affiliated institutes international collaborations and region on citations received Sin (2011); Yan and Ding (2012) by them. Studying both the networks together also helps in forming ranking measures for institutes and researchers Xu et al. (2017); Zhan and Tse (2017) in scientific collaboration.
Simultaneous analyses of citation and co-authorship networks have given insights into correlations between the two networks. Considering the effect of one network on the other gives a better understanding of the true nature of scientific collaborations. While earlier studies have addressed citation and co-authorship networks simultaneously and established a strong interdependence between the two, there is still scope to understand the details of correlation between these networks. In this study we build on the strong correlation between interacting pairs of authors in citation and co-authorship networks. First, we define a method which tracks the evolution of relationships between each possible author–author pair in both networks. Next, we formulate a null hypothesis for the probability of citation exchange between a pair of authors and use probabilistic analysis to compare it with empirical observation from networks constructed using the publications in the American Physical Society (APS) journals between 1970-2013. This way, we capture both macroscopic and microscopic changes in network structure and address a number of questions which are otherwise difficult to answer.
What fraction of authors exchange citations but do not co-author however are connected in the co-authorship network?
How are citations exchanged between co-authors?
How do the statistics in 1 vary with network distance between authors?
How does receiving a new citation affect the likelihood of an author creating a new link in the co-authorship or the citation network?
What is the relationship between the probability of citations and network distance?
What is the waiting time distribution for consecutive co-authorship events and for consecutive co-citation events?
Our analysis is based on a similar approach used by earlier studies Barabási et al. (2002); Biscaro and Giupponi (2014); Chen et al. (2013); Kas et al. (2012); Martin et al. (2013); Yan and Ding (2012). However, our method is significantly different from theirs. By tracing the interactions between all possible pairs of authors in the citation network and at all existing shortest path lengths in the associated co-authorship network, we are able to see in detail the effect of collaborations on citations and vice-versa. Our main contribution is to explain the effect of the distance in the co-authorship network on the citations exchanged between pairs of authors.
The remainder of this paper is organized as follows: Sec. II presents our methods and explains how we create the distance and citation matrices, as well as how we perform our empirical calculations. In Sec. III, we present and discuss our results. We summarize our findings and possible extension of this work in Sec. IV.
For the purpose of our study, we construct a longitudinal data set of publications by Indian researchers in the American Physical Society (APS) journals between 1970 and 2013. Here we consider an author to be Indian if they have any paper with an Indian affiliation. Therefore all papers with authors having at least one Indian affiliation are included in the data set. There were 14,703 such papers Singh and Jolad (2019). For the extracted papers we performed name disambiguation on the authors names to assign a unique ID for every author. This was done to account for different naming styles used by authors over time. For naming disambiguation we use edit distance between strings to cluster similar names and then check for neighborhood overlap in the co-authorship network. Names with small edit distance and high neighborhood overlap were grouped together and manually checked for uniqueness using information from online databases. This results in 8,084 unique Indian authors.
Then, we construct bipartite graphs for every year, from 1970 to 2013, where is the set of papers, is the set of authors and is the set of edges connecting nodes from and . Each graph at time is cumulative, storing all the information until time . From each we generated weighted, cumulative, and undirected projected co-authorship networks. We also construct cumulative directed citation networks for every year using the paper IDs of Indian publications from the APS citations data set. We illustrate the process of creating theses networks in Fig. 1. The ordering of node labels in the projected co-authorship networks is kept consistent for all calculations. Using the above graphs we construct our data matrices for co-authorship and citation networks as follows.
ii.1 Data Matrices
In order to aid our analysis, we created two types of matrices, one for the co-authorship networks and the other for the citation networks, for each of the 44 years. The matrices have size , where is the number of unique authors in the whole data set.
The elements of the first matrix type, , are given by
such that the matrix captures the distances between all possible pairs of authors in the network at time . For , it can mean that the nodes do not exist in the network at that time or that they are not connected via any path.
The second matrix type, , stores the citations exchanged between papers written by and until a given time . That is, is the cumulative number of times that cites until that particular year.
To see the aging effect in collaboration between authors, we trace the history of co-authorship events. First, we extract the distance collaborations at the end of the time period (i.e. edges in the 2013 co-authorship graph). Next, we trace the presence of edge between and (contributing authors) in reverse order. The time when the edge first appears is marked , which is the year of first collaboration. Then, we check for the presence of and in the network before until we find , the first year in which authors and both are present in the network. For all these years, the number of citations exchanged by pairs of authors, at every time step before and after collaboration, are recorded. A diagram illustrates this method of tracing the history of collaboration between pair of nodes in Fig. 2.
In order to compare the history of citation and co-authorship for all pair of authors, we shift the time series on the x-axis of every pair to zero. That is, we adjust , the year of first collaboration between two authors, to 0. Out of these, we remove the pairs that had . We do this to remove authors who appear in the network together with a shared publication. As these authors will not have any citing history prior to their first collaboration, we exclude them. The remaining are pairs of authors who took at least one year to collaborate after appearing together in the network.
The data matrices and store the information of the distance and the citations exchanged between all possible pairs of nodes for the co-authorship and citation networks, respectively, for every time step (which is an year in our case). This enables us to calculate any changes in distance or citations exchanged from one year to another. Then, to address the research questions mentioned in the Introduction, we define our calculations based on the different situations that each possible pair of nodes present in the networks. More specifically, we calculate the following, at every time step :
What fraction of authors exchange citations but do not co-author?
We count pairs that exchange citations () and have a connected path in the co-authorship network ().
We count pairs that exchange citations () but are not connected in the co-authorship network . These are the pairs that are aware of each others work via citations, but do not have any direct or indirect connection with each other via co-authorship.
How are citations exchanged between co-authors?
We count pairs that co-author () but do not exchange citations .
We count pairs who are co-authors () and exchange citations ().
How do the statistics in 1 vary with network distances between authors? We count pairs that exchange citations () for different distances .
How does receiving a new citation affect the likelihood of an author creating a new link in the co-authorship or the citation network?
Response by authors, in terms of citations (how they cite back), to other authors who cited them, and to the total citations they received. For every author in the citation matrix , we define as the number of authors that cite . Among , is the number of authors whom cites back. Similarly, are the total citations received by from the set of authors and are the total number of citation given out by to the set of authors. The response by an author is calculated as: a) — the response to citing authors; and b) — the response to citations received.
What is the relationship between the probability of citations and network distance ?
Correlation between citations exchanged and network distance. In order to define the probability of citation between author, we take as the probability that cites at time , given by
where is the fraction of citation has received prior to , and is the fraction of citations gives out in the year between and . We also define the mean probability
We calculate and from our data matrices according to
Our reasons behind this approach are twofold: (i) Popular authors (or papers) have a greater tendency to get cited () — the frequently observed preferential attachment phenomenon; and (ii) if an author (paper) is giving out more citations it uniformly increases the probability of other authors (papers) getting cited (
). It should be noted that this definition is independent of the relationship between authors in the co-authorship network. This will serve as the null model for our subsequent comparisons since it calculates probability distributions for citations without accounting for co-authorship distance.
First, Eq. (1
) estimates the probability of a directed edge between authors in the citation graph, given no other information than the number of citations exchanged between authors recorded in
. Next, we need to define the relations to compare the empirical observations from the citation and co-authorship networks with the null model. We use Bayes’ theorem to construct probability relationships. This helps us to put constraints in our observations that will help us to highlight the dependency of citations on the shortest path between authors in the co-authorship network.
Then, if is the time of first co-appearance of and and is the probability of them co-authoring, we have
Eq. (6) is the empirical probability of observing pairs of authors connected by a path of length one, given that they exchange non-zero citations (), normalized by the probability of non-zero citations for pairs at all possible shortest path lengths (Eq. (7)). Therefore for any distance , we have
If we reverse the relationship using Bayes’ rule, the empirical probability of observing pairs with non-zero citations between them, given they are at distance is calculated according to
The denominator in Eq. (9) normalizes over pairs that either exchange citations () cite or do not cite each other () given a network distance .
What is the waiting time distribution for consecutive co-authorship events and for consecutive co-citation events?
Co-authorship events. The co-authorship networks are weighted undirected networks constructed cumulatively. Therefore, every time an author shares a paper with another one, the weight of the edge between them in the co-authorship network changes. We record the time it takes for this change to happen. We do so for all pairs of authors in the network over time.
Co-citation events. For every pair in the citation matrix we record the time it takes for a change in the value of over the whole time period.
With our data matrices, and , and the distinct relations between citations exchanged and distance in the co-authorship networks defined, we now turn our focus to the understanding of the interdependence between the co-authorship and the citation networks. That is what we address in the next section.
Iii Results and Discussions
Using the data matrices and , described above, we count the number of pairs for different citation and co-authorship distance relations as the networks evolve with time. To understand the interdependence between the citation network and its associated co-authorship network we calculate the citations exchanged between pairs by splitting them into three groups:
For each group we separately count the number of pairs of authors that do not exchange any citation (orange line in Fig. 3) and pairs that have non-zero citations shared between themselves (blue line in Fig. 3). For all cases, as the citation and co-authorship networks grow, the fraction of pairs that do not have any citations between them is larger than the the fraction of pairs that do exchange citations.
The contributions of each of the three groups defined to the total number of citations are shown in Fig. 4. First, the light green region in Fig. 4 is the fraction of citations exchanged between pairs that have distance in the co-authorship networks. Even though the number of such pairs is a small fraction in the co-authorship networks (blue line in Fig. 3 (a)) they still contribute significantly to the total number of citations. Second, pairs that co-author are responsible for most of the citations (blue region in Fig. 4), which reflects the importance of an authors’ collaborators to the number of citations received. And third, disconnected pairs exchange a very small fraction of total citations between them (sky blue region in Fig. 4), showing a decreasing trend, until it almost vanishes in the last years.
The behavior described above could be a consequence of our choice of the data set. Since we focus on a small fraction of the total number of publications (those from Indian authors) in the global APS network, authors are expected to be well connected and closely citing each other. As most of the pairs are connected by a path in a growing co-authorship network — represented by the orange line in Fig. 3(a) approaching one — and are aware of each other in the network (the decrease in the orange line in Fig. 3(c)) the citations by distant pairs decreases. The trend in the number of co-author pairs that exchange citations shows an interesting sudden jump in the mid 1990s. We believe this trend is due to the increasing number of nodes in the co-authorship network. In the beginning there were very few nodes (authors) in the network, most of whom appeared as co-author pairs which explains the initial increase. Post 1993 (the blue line in Fig. 3(b)) we notice a sudden increase in the number of such pairs. This sudden change is because of the introduction of papers with a high number of authors in that period. These papers lead to large cliques (totally connected sub-graphs) in the co-authorship network. Most of these papers are published by large collaboration groups often having multiple common authors in their publications and with many Indian authors being part of such groups. For example, the papers Abachi et al. (1994, 1995); Abelev et al. (2010) have 351, 395, and 383 authors respectively. Even one citation shared between such papers would dramatically inflate the number of citations exchanged, due to the large size of the induced co-author cliques.
In the above calculations, we counted the total citations exchanged between the pairs in the co-authorship network for different network distances normalized by the total number of possible pairs in the co-authorship network. The citation count included both incoming and outgoing citations. To measure the response of authors to incoming citations we split our calculations in two parts. First, we calculate the average fraction of out-going citations from authors for every citation received by them. We observe that over time people tend to cite more and more articles in their work hence, we see an increasing trend in response to citations (the orange curve in Fig. 5). Second, we calculate the average fraction of incoming citations that an author responds to by subsequently citing the author who initially cited her (the blue curve in Fig. 5). When the co-authorship network is in its initial phase, with a small number of researchers, most co-author pairs cite each other. Hence, we observe high citation reciprocity at initial times. As the network grows, the distribution of citations becomes more heterogeneous as some authors receive more citations than others (authors of influential papers receive many citations). In addition, authors that are no longer publishing cannot reciprocate anymore but still receive citations. Thus, more citations are given out than received in average, which results in a decreasing trend in citation reciprocity (the blue line in Fig. 5). The sudden increase in reciprocity in the mid 1990s is due to the citations exchanged between papers with a large number of authors, which, as aforementioned, started to appear around that time.
So far, our observations give a macroscopic understanding of the interdependence of simultaneously growing citation and co-authorship networks. To probe that further, we make more elaborate calculations to see the effect of co-authorship network distance on the citations exchanged. In the interest of this objective we ignore pairs that do not exchange any citations (orange line in Fig. 3) from our subsequent analysis.
The number of citations exchanged by pairs of authors at distance in the co-authorship network decreases rapidly with increasing co-authorship distance (Fig. 6). We plot this relation for networks at different times (1990, 2000, and 2013) to show that the trend is consistent as the network evolves. The average citations between pairs (Fig. 7(a)) displays significant difference in the temporal trends for different network distances. For direct collaborations (pairs with ), the rate of citation exchange increases with time, more rapidly after 1995. This is likely due to a sudden increase in the number of collaborators (as seen by the change in the average degree of nodes in Fig. 7(b)). The average number of citations differ roughly by an order of magnitude for co-authorship distances of ; for larger distances (), the average number of citations exchanged are very low, with similar trends, as show in Figs. 7(a) and 7(a) (inset).
The change in the strength of collaboration between two authors and is measured by the total citations exchanged between and (scatter plot Fig. 8(a)) before and after the period of first collaboration (black line in Fig. 8(a)). This includes cases where and cite their own previous papers. Co-author pairs exhibit an interesting citing pattern — the number of citations shows a steep rise after the first co-authorship event and then decays with time, indicating an aging effect.
Interestingly, the peak for this distribution is within five years of . When averaged over all times, we notice that the decaying trend is well fitted by a Weibull distribution, , as was noted in Börner et al. (2004).
Next, we calculate the waiting time () distribution for consecutive citations between pairs of authors and for consecutive co-authorship events. Both follow a similar trend (Fig. 9) with the majority of co-authorship and citation events () happening within the first five years of the initial event (black dotted line in Fig. 9).
Finally, we calculate the empirical probabilities of authors acquiring citations, firstly with the null model (Eq. (1)) and subsequently with the empirical probabilities derived using Bayes’ formalism (Eqs. (6-9)). The latter accounts for co-authorship distance in determining the probability of citations exchanged between pairs (Fig. 10). We notice that the null model, which is proportional to popularity of the author (paper), and the number of citations given out by the citing author (paper) are not sufficient to explain the observed behavior of citation exchange. The citing patterns significantly differ for different network distances between author pairs. The probability of citations between pairs at (blue line in Fig. 10) closely follows the null model, while pairs at greater distances significantly differ from it. From this we infer that most of the overall citation behaviour is explained by only considering those citations that come directly from co-authors with . This is also made evident by the blue region in Fig. 4 where citations from co-authors contribute to most of the total number of citations. Distinct probabilities of citations at different co-authorship distances (Fig. 10) and the decay in average citations with higher network distances (Fig. 6) indicate an interdependence between citing patterns and co-authorship network distances, hence confirming our hypothesis.
By splitting our analysis into different research questions, we were able to explore both macroscopic and microscopic trends in citing patterns between authors as they appear in the associated co-authorship network.
Very small fraction of pairs are connected with distance in the co-authorship network but still have a significant contribution to overall number of citations.
Co-authors are a very small fraction of the total possible number of pairs but account for most of the citation exchanges observed.
Disconnected pairs contribute to citations in the beginning of the network, but their contribution becomes negligible as the network grows and becomes more connected.
Besides interactions between pairs, we also investigated the average probability of an author citing back an author that has cited her (the blue line in Fig. 5) and the average probability of an author giving back a citation for every citation received (the orange line in Fig. 5). The trends indicate that while, over time, the ratio of outgoing to incoming citations per author has increased. However, the ratio of pairwise outgoing citations that reciprocate citing authors decreases over time. In Fig. 8, the plot of citations exchanged between authors exhibits a sudden increase when they co-author a paper and then decays, which is consistent with the aging effect in citations and collaboration reported by earlier studies. A similar effect is observed in Fig. 9 where of consecutive co-authorship and citation events happen within the first five years of an initial co-authorship event.
On the other hand, in microscopic calculations, we first calculate the average citations shared between author pairs at all possible network distances for networks at different points in time (Fig. 6) and for all time steps (Fig. 7 — plotted only up to for clarity). The number of average citations shows a steep decay up to and then is almost stable for longer distances. That is, pairs that are more than distance three apart in the co-authorship network have a similar (and minimal) effect on citation patterns. To confirm the interdependence between citation networks and the associated co-authorship network we formulate a null model for the probability of an author citing author in Eq. (1) and then using Bayes’ formalism (Eq. (6-9)) we explicitly show that the null model is indeed insufficient to explain the citing patterns. There is a significant effect caused by the distance between pairs of authors in the co-authorship network over their citing patterns. The effect being most dominant for immediate co-authors (Fig. 10).
The main contribution of this study lies in a rigorous and a comprehensive analysis that probes the relations between the distances in the co-authorship network and the citation patterns in the citation network. We do this for all possible pairs of researchers from the citation and co-authorship networks constructed. For all pairs of authors, we observe the number of citations exchanged between them as a function of distance in the co-authorship network, as it evolves over time. We find that co-authors dominate the citation patterns in our networks. The remainder of the citations were mostly between pairs that have a short () connected path in the co-authorship network; the average number of citations exchanged decaying with increasing distance. Pairs with distances have a relatively small contribution to the citations. Disconnected pairs of authors make a small contribution to the citations in the initial years of the network but quickly become almost negligible, as the network grows to be more connected over time. We also highlight the underlying aging effect in mutual citations and collaborations.
Most of the citations are accounted for within three degrees of separation in our data set. This indicates that authors mostly cite their co-authors and the collaborators of their co-authors. Since we study researchers affiliated to Indian institutes these citations exchanged would be between authors working in similar research topics with a likely connection in the co-authorship network. In short similarity in research topic and affinity towards close collaborators would be the major effects driving the citation patterns in our case. However as some authors (or papers) gain more citations over time the dynamics for the top cited authors (papers) are likely to differ from those of the majority. In such cases, distant or disconnected authors will also have a significant contribution to the total number of citations of the author (or paper). Therefore to measure the impact of a paper with respect to citations received we should have measures that account for this possible bifurcation in citing patterns.
The main advantage of our data set was in calculating pairwise interactions. Even so, we realize that our data set is not completely comprehensive, as it considers only Indian authors and APS publications, hence our results might show small variations when calculated for larger data sets. However, we believe that this difference should not be considerably large, assuming that pairwise interactions between authors would be similar irrespective of the size of the network.
We reflect on the opinion that most real-world networks can be viewed as interdependent multi-layer networks, with networks of scientific collaborations as an example of this. This interdependence is critical when studying the dynamics of such networks. Our analysis explicitly shows that connected paths in one network (co-authorship) impact the structure of the other (citation). Dominance of pairs close in distance highlight the importance of an authors’ neighborhood in her citing patterns which, in turn, can be used to explain the patterns in flow of ideas and information in the scientific ecosystem. Results from this study can be used in the development of more sophisticated models to investigate the spread of scientific knowledge. We believe that to understand the true nature of research collaboration, it is important to consider both co-authorship and citation networks simultaneously.
The authors would like to thank Anand Sengupta, Anirban Dasgupta, Krishna Kanti Dey, Jayesh Choudhary and Amit Reza for their valuable insights and discussions during the period of this study.
- Newman (2001a) M. E. J. Newman. The structure of scientific collaboration networks. Proc. Natl. Acad. Sci. USA, 98(2):404–409, 2001a.
- Newman (2001b) M. E. J. Newman. Scientific collaboration networks. ii. shortest paths, weighted networks, and centrality. Phys. Rev. E, 64(1):016132, 2001b.
- Sinatra et al. (2016) R. Sinatra, D. Wang, P. Deville, C. Song, and A.-L. Barabási. Quantifying the evolution of individual scientific impact. Science, 354(6312):aaf5239, 2016.
- Dong et al. (2017) Y. Dong, H. Ma, Z. Shen, and K. Wang. A century of science: Globalization of scientific collaborations, citations, and innovations. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1437–1446. ACM, 2017.
- Tomassini and Luthi (2007) M. Tomassini and L. Luthi. Empirical analysis of the evolution of a scientific collaboration network. Physica A, 385(2):750–764, 2007.
- Newman (2004) M. E. J. Newman. Coauthorship networks and patterns of scientific collaboration. Proc. Natl. Acad. Sci. USA, 101(suppl 1):5200–5205, 2004.
- Huang et al. (2008) J. Huang, Z. Zhuang, J. Li, and C. L. Giles. Collaboration over time: characterizing and modeling network evolution. In Proceedings of the 2008 international conference on web search and data mining, pages 107–116. ACM, 2008.
- Martin et al. (2013) T. Martin, B. Ball, B. Karrer, and M. E. J. Newman. Coauthorship and citation patterns in the physical review. Phys. Rev. E, 88(1):012814, 2013.
- Vasques Filho (2018) D. Vasques Filho. Structure and dynamics of social bipartite and projected networks. PhD thesis, The University of Auckland, 2018.
- Barabási et al. (2002) A.-L. Barabási, H. Jeong, Z. Néda, E. Ravasz, A. Schubert, and T. Vicsek. Evolution of the social network of scientific collaborations. Physica A, 311(3-4):590–614, 2002.
- Chen et al. (2013) Y. Chen, K. Börner, and S. Fang. Evolving collaboration networks in scientometrics in 1978–2010: a micro–macro analysis. Scientometrics, 95(3):1051–1070, 2013.
- Gualdi et al. (2011) S. Gualdi, C. H. Yeung, and Y.-C. Zhang. Tracing the evolution of physics on the backbone of citation networks. Phys. Rev. E, 84(4):046104, 2011.
- Shi et al. (2009) X. Shi, B. Tseng, and L. Adamic. Information diffusion in computer science citation networks. In Third International AAAI Conference on Weblogs and Social Media, 2009.
- Pan et al. (2018) R. K. Pan, A. M Petersen, F. Pammolli, and S. Fortunato. The memory of science: Inflation, myopia, and the knowledge network. J. Informetr., 12(3):656–678, 2018.
- Higham et al. (2017) K. W. Higham, M. Governale, A. B. Jaffe, and U. Zülicke. Unraveling the dynamics of growth, aging and inflation for citations to scientific articles from specific research fields. J. Informetr., 11(4):1190–1200, 2017.
- Börner et al. (2004) K. Börner, J. T. Maru, and R. L. Goldstone. The simultaneous evolution of author and paper networks. Proc. Natl. Acad. Sci. USA, 101(suppl 1):5266–5273, 2004.
- Fiala et al. (2015) D. Fiala, G. Tutoky, P. Koncz, and J. Paralič. Ageing of edges in collaboration networks and its effect on author rankings. Acta Polytechnica Hungarica, 12(6):149–160, 2015.
- Tutoky and Paralič (2011) G. Tutoky and J. Paralič. Time based modeling of collaboration social networks. In International Conference on Computational Collective Intelligence, pages 409–418. Springer, 2011.
- Hajra and Sen (2005) K. B. Hajra and P. Sen. Aging in citation networks. Physica A, 346(1-2):44–48, 2005.
- Kas et al. (2012) M. Kas, K. M. Carley, and L. R. Carley. Trends in science networks: understanding structures and statistics of scientific networks. SNAM, 2(2):169–187, 2012.
- Keegan et al. (2013) B. Keegan, D. Horn, T. A. Finholt, J. Kaye, et al. Structure and dynamics of coauthorship, citation, and impact within cscw. arXiv preprint arXiv:1307.7172, 2013.
- Amblard et al. (2011) F. Amblard, A. Casteigts, P. Flocchini, W. Quattrociocchi, and N. Santoro. On the temporal analysis of scientific network evolution. In 2011 International Conference on Computational Aspects of Social Networks (CASoN), pages 169–174. IEEE, 2011.
- Ding (2011) Y. Ding. Scientific collaboration and endorsement: Network analysis of coauthorship and citation networks. J. Informetr., 5(1):187–203, 2011.
- Tol (2011) R. S. J. Tol. Credit where credit’s due: accounting for co-authorship in citation counts. Scientometrics, 89(1):291, 2011.
- Glänzel and Thijs (2004) W. Glänzel and B. Thijs. Does co-authorship inflate the share of self-citations? Scientometrics, 61(3):395–404, 2004.
- Wallace et al. (2012) M. L. Wallace, V. Larivière, and Y. Gingras. A small world of citations? the influence of collaboration networks on citation practices. PloS one, 7(3):e33339, 2012.
- González-Teruel et al. (2015) A. González-Teruel, G. González-Alcaide, M. Barrios, and M. Abad-García. Mapping recent information behavior research: an analysis of co-authorship and co-citation networks. Scientometrics, 103(2):687–705, 2015.
- Biscaro and Giupponi (2014) C. Biscaro and C. Giupponi. Co-authorship and bibliographic coupling network effects on citations. PLoS One, 9(6):e99502, 2014.
Sarigöl et al. (2014)
E. Sarigöl, R. Pfitzner, I. Scholtes, A. Garas, and F. Schweitzer.
Predicting scientific success based on coauthorship networks.
EPJ Data Science, 3(1):9, 2014.
- Guo et al. (2017) L. Guo, X. Cai, F. Hao, D. Mu, C. Fang, and L. Yang. Exploiting fine-grained co-authorship for personalized citation recommendation. IEEE Access, 5:12714–12725, 2017.
- Lande and Andrushchenko (2016) D. V. Lande and V. B. Andrushchenko. Formation of subject area and the co-authors network by sounding of google scholar citations service. arXiv preprint arXiv:1605.02215, 2016.
- Sin (2011) S.-C. J. Sin. International coauthorship and citation impact: A bibliometric study of six lis journals, 1980–2008. J. Am. Soc. Inf. Sci. Tec., 62(9):1770–1783, 2011.
- Yan and Ding (2012) E. Yan and Y. Ding. Scholarly network similarities: How bibliographic coupling networks, citation networks, cocitation networks, topical networks, coauthorship networks, and coword networks relate to each other. J. Am. Soc. Inf. Sci. Tec., 63(7):1313–1326, 2012.
- Xu et al. (2017) G. Xu, X. He, and A. Butler. What comes first the co-authorship network or the citation? In ACM International Conference in Proceedings Series, 2017.
- Zhan and Tse (2017) C. Zhan and Chi K. Tse. A network model for growth of publications and citations. Journal of Complex Networks, 5(2):303–314, 2017.
- Singh and Jolad (2019) C. K. Singh and S. Jolad. Structure and evolution of indian physics co-authorship networks. Scientometrics, 118(2):385–406, 2019.
- Abachi et al. (1994) S. Abachi et al. First generation leptoquark search in pp¯ collisions at√ s= 1.8 tev. Phys. Rev. Lett., 72(7):965, 1994.
- Abachi et al. (1995) S. Abachi et al. Top quark search with the d0 1992–1993 data sample. Phys. Rev. D, 52(9):4877, 1995.
- Abelev et al. (2010) B. I. Abelev et al. Identified particle production, azimuthal anisotropy, and interferometry measurements in au+ au collisions at s nn= 9. 2 gev. Phys. Rev. C, 81(2):024911, 2010.