• Bojanowski M., Chroł B. (2018) “Proximity-based Methods for Link Prediction in Graphs with R package ‘linkprediction'” Preprint.

Link prediction is a problem of predicting future edges of an undirected graph based on a single snapshot data of that graph. Vertex proximity measures are indicies giving numerical scores for every pair of vertices in a graph that can be used for predicting future edges. This short note describes an R package linkprediction implementing 20 different vertex similarity and proximity measures from the literature. The article provides the definitions of implemented measures, describes the main user-facing functions, and illustrates the use of the methods with a problem of predicting future co-authorship relations between researchers of University of Warsaw.

On some fundamental level, we can think of scholars as actors possessing, or controlling, various types of resources. Collaboration in science is understood here as a process of pooling and exchanging such resources. We show how diversity of resources engaged in scientific collaboration is related to the structure of collaboration networks. We demonstrate that scholars within their personal networks simultaneously (1) diversify resources in collaboration ties surrounded by structural holes and (2) specialize resources in collaboration ties embedded in dense collaboration groups. These complementary mechanisms decrease individual efforts required to maintain effective collaborations in complex social settings. To this end, we develop a concept of “pairwise redundancy” capturing structural redundancy of ego’s neighbors vis-à-vis each other.

  • Czerniawska D., Fenrich W., Bojanowski M. (2018) “Actors, Relations and Networks: Scholarly Collaboration Beyond Bibliometric Measures” Polish Sociological Review 2(202):167-185

Scholarly collaboration is relatively well described quantitatively on the macro level thanks to the analyses of large bibliographic databases. At the same time, there are known  limitations of the bibliometric approaches to studying collaboration in science. We argue that in order to improve our understanding of social processes operating in science it is necessary to take a more in-depth look: (1) identify kinds of actors that are recognized as potential partners in collaboration, (2) what features of collaborative relations are considered crucial for engaged actors, (3) what kinds of structures of networks composed of collaboration relations actors are embedded in, and what factors influence these structures. With 30 individual in-depth interviews (IDI) with Polish scholars we gathered detailed information about individual collaborations that allowed us to analyze ollaborative ties from individual perspective and map respondent-centered networks of collaboration. Scholars identify individuals as well as teams or institutions as collaborators. They also distinguish symmetric and asymmetric collaborations. Structures of respondent-centered collaboration networks are affected by (a) leadership strategies of team principals (especially whether teams are built around positions or individuals); (b) institutional location (by making establishing external collaborations easier for scientists from bigger institutions); (c) scientific degree and recent changes in financing of science (as young scientists receive more freedom from usual organizational hierarchies by receiving substantial grants).

  • Czerniawska D., Bojanowski M. (2017) “Are the ‘big ponds’ in the core? Overlap of tyhe core/periphery collaboration structure and ‘big pond’/’small pond’ segregation. Presentation at EUSN2017 conference. September 26-29, Mainz, Germany.

The main aim of the paper is verifying to what extend the core/periphery collaboration structure in science overlaps with ‘big pond’ vs. ‘small pond’ segregation. Both of these concepts are widely applied in social network in science. We expect that ‘big ponds’ with resources they control will attract more productive scientists, which will result in more dense collaboration networks affiliated with ‘big ponds’. The analysis is conducted based on Polish scientific community. Publication database covers years 2000-2015 with 200 000 scientists and 400 000 publications. ‘Big pond’/’small pond’ classification is based on the size of institution operationalised as total amount of research funds obtained for fundamental research, a number of active scientists and government evaluation conducted every four years.

  • Bojanowski, M. (2017) Networks of co-production in science. Presentation at ARS`17 conference, May 17, Naples, Italy.

Collaboration in science takes many forms. Some of these forms, such as providing substantive research input, can be understood as contributing resources to collective “projects”. We propose a mathematical model of intellectual collaboration that is based on the following assumptions: (1) scholars do research and write scientific publications; (2) each publication consists of a finite set of research components/pieces; (3) scholars are characterized by a finite set of skills that enable them to create above mentioned components, but with varied effectiveness; (4) scholars form collaboration ties to pool their skills and work more effectively. We analyze the model analytically and with computational methods. We discuss implications for actor-oriented modeling of dynamics of collaboration networks.

  • Czerniawska D. Bojanowski M. (2017) Structural holes and tie redundancy in scientific exchange networks. Presentation at ARS`17, May 17, Naples, Italy.

On some fundamental level, we can think of scholars as actors possessing, or controlling,  arious types of resources. Engaging in scientific collaboration requires a lot of time and energy to provide resources needed to reach desired outcome. In principle we should assume that in scientific collaborations scholars combine resources they control and share the outcomes according to agreed terms in contrary to passing resources from one to another. The more diverse resources a scholar is engaging into a collaboration, the more time- and energy-consuming a collaboration becomes. In order to reduce the engagement a scholar may decide to invite an additional collaborator to delegate the responsibility for providing some of the resources directly to herself and to other common collaborators. We argue that such process should lead to higher specialization levels in densely connected subgroups (more redundant ties, less opportunities to bridge structural holes). The study is based on 40 individual in-depth interviews. Data contains information about collaboration ego networks and resources engaged in collaboration by egos and alters. Resources are coded into several large categories like knowledge and skills, funding, connections. Every ego-alter relation is described with sets of resources engaged by each party. We explain the diversity of resources engaged in different collaborations with the extent of  structural redundancy between corresponding collaboration ties. The measure of  redundancy is based on the connectivity and embeddedness of alters being compared (geodesic distance and number of shared partners after removing ego-alter ties).

The main aim of the paper is discussing the exchange processes in collaboration ego-networks among scientists. On some fundamental level, we can think of scholars as actors possessing, or controlling, various types of resources. These resources can be roughly grouped into the following categories: human capital resources including skills and knowledge; social capital resources including social status and social connections to other researchers; financial capital resources including access to and control of research funds. Desirability and uneven distribution of these resources between different scholars create opportunities for collaboration that take the form of exchange. Previous research has developed general rules for exchange: behaviour is motivated by the desire to increase gain and to avoid loss, exchange relations develop in structures of mutual dependence, actors engage in recurrent, mutually contingent exchanges with specific partners over time, valued outcomes obey the economic law of diminishing marginal utility. However, it does not take into consideration types of resources, which are substantial for understanding scientific collaboration networks. Based on 30 IDI conducted with Polish scholars we show what resources are a subject of exchange; what are the motivation to initiate and engage in exchange; what are the norms regulating the exchange of different types of resources?

Świat nauki, naukowcy oraz charakter pracy naukowej składają się na fascynujący system społeczny, w którego funkcjonowaniu różne normy, hierarchie organizacyjne i statusowe mieszają się z konkurencją i współpracą. Szczególnie współpraca pomiędzy naukowcami wydaje się mieć istotne znaczenie jako mechanizm pozwalający na łączne korzystanie z różnego rodzaju wyspecjalizowanych zasobów. Sieci istniejącej współpracy stanowią tzw. “Niewidzialny(-e) Uniwersytet(y)”. Przedstawimy wybrane wyniki realizowanego aktualnie projektu dotyczącego konkurencji i współpracy w nauce polskiej, jak odkrywać “Niewidzialny Uniwersytet” i jak wygląda on w Polsce oraz jakich istotnych zjawisk badając “Niewidzialny Uniwersytet” tradycyjnymi metodami nie poznamy.

Abstract: In recent decades doing science has become more and more a collective endeavor requiring cooperation crossing institutional and disciplinary boundaries. Although this phenomenon has been defined and described on macro level, the knowledge about individual motivations, norms and values fostering scholarly communication is limited.Evidence brought by scientometric studies misses out cooperation which is not related to co-authorship. Moreover, since scholarly community varies greatly in co-authorship norms, this kind of information have different meaning varying from field to field, especially in the case of Poland, where recent reforms of legal and financial framework were meant to encourage researchers to intensify cooperation within and outside academia. Questions like: “How does scientific collaboration is initiated?”, “What are the motivations to engage in new collaborations?” or “Under what conditions do collaborations last in time?”, demand different, qualitative tools. Our study, based on 30 IDI’s with Polish scholars representing various disciplines and research centers of different quality and size, revealed the existence of important differences in initiating and conducting teamwork activities. Teams are built in different ways: by seeking a new collaborator with rare set of skills, reviving existing but temporarily inactive network of researchers known from earlier projects, or simply engaging in-house subordinates. Motivations and rewards important to team members playing particular roles also differ significantly. Gathered qualitative material will help interpreting the results of future quantitiative studies and inspire new hypotheses that can be tested using large datasets.

Although interdisciplinary research has been present in science and the humanities for decades, only recently has it become more and more common in an everyday scholars’ experience. The general shift towards interdisciplinarity was captured in the mid 90’s (Gibbons, 1994), but lately it has been increasingly fostered by individual, technical, and institutional factors. Researchers might perceive interdisciplinarity as a tool to look for new questions or new methods. The rapid growth of technology offers novel ways to gather, compute and analyse data, which are applied in many disciplines. The funding agencies see interdisciplinary research as a way to face the most challenging issues of our time. On the other hand, engaging in interdisciplinary research might imply a wide range of difficulties from a team design to getting funding and a proper outcome evaluation. These conditions might influence individual decisions on if and how to undertake this path. The origins of motivations and incentives (individual vs. institutional) will be tackled in this presentation as the innate, ideational aspects of interdisciplinarity. The other crucial aspect is how the ability to mobilize different types of resources (like skills, expertise, infrastructure) throughout a professional network influence the decision to engage in interdisciplinary research. These two problems will be addressed with data from 30 individual in depth interviews conducted last year among scholars coming from variety of disciplines. The individual stories will allow to reconstruct the types of personal networks, including human and non-human actors and their role in the process of engaging in and conducting interdisciplinary research.

The scholarly collaboration phenomena is relatively well described on macro level, thanks to network analyses based on large bibliographic datasets. Nevertheless, qualitative in depth knowledge about the nature of scholarly collaboration networks going beyond publication matters is scarce. Also the innate, ideational aspects of the phenomena in question remain invisible from this wider perspective. Based on the 30 individual in-depth interviews with Polish scholars we were able to sketch 30 ego-centered networks representing his or her immediate collaborations. Respondents were asked questions like: “Who would you indicate as your collaborator?”, “Do you collaborate with people from other institutions/academic centers/countries?” “How did your collaboration start?” and had an opportunity to freely describe their professional environment. As a result, we were able to outline values, norms and roles standing behind particular networks. Collaborations are mostly in-house, and occur within small research teams supported by relatively few outsiders: students, PhD candidates, visiting scholars, or researchers from outside the institution. We can indicate some factors influencing the structure and dynamics of collaboration networks like the propensity to limit the area of research interests. The sequential design of the research project enables utilization of its initial part to direct future quantitative analyses and improved qualitative studies.

We tackle the problem of identifying pairs of disconnected actors that are likely to form a link in the future. We compare two approaches: (1) several node-proximity methods, based on overlapping neighborhoods, random walks on graph and distances between actors, and (2) estimation of Exponential Random Graph Models, which allows to calculate model-based conditional probability of tie existence. In all cases we select edges, which are the most probable, and compare it to true values. In addition we counted Area Under Curve Measure to asses general performance. The best proximity based method turns out to be the Random Walk with Restart. It performs significantly better than the simplest methods. As for ERGM, the best model consists of following terms: number of edges, number of edges where both nodes have the same affiliation and number of nodes with given degree. ERGM turns out to perform worse than proximity-based methods.

  • Bojanowski M. (2014) Individual Publication Strategies and Collaboration in Authorship: Analysis of Co-authorship Network from a Large University. Presentation at Sunbelt XXXIV conference.

Publications are the main medium of scientific communication. They are the primary way of documenting and communicating research results. Different scientific disciplines developed different norms regarding the utility and function of different types of publications, i.e., journal articles, books, edited volumes, or conference proceedings, etc. Choice of a particular publication types for a stream of individual publications is an element of a broader individual publication strategy. The other element of that strategy is whether these publications, and research they document, are produced in collaboration with others. Scientometric studies report a steadily growing number of co-authored research articles. Co-authorship is an indicator of social relations between scientists: collaboration, but often also authority. Scientific disciplines differ in terms of norms related to co-authorship too. We present a dynamic analysis of co- authorship between the researchers from a large university. The data span 10 years and can be considered complete: contain all relevant publications of about 20+ thousands employees of the university. We investigate dynamic patterns and disciplinary differences in publication strategies in terms of both the structure of co-authorship networks and types of publication.