Gephi and Its Context: Three Oppositions in The Epistemic Culture of Network Analysis And Visualisation

(1)

University of Amsterdam

RMA Media Studies

Master’s Thesis

GEPHI AND ITS CONTEXT

Three Oppositions in The Epistemic Culture of Network Analysis And Visualisation

Emilija Jokubauskaite

Supervised by: dr. Bernhard Rieder

Second reader: dr. Thomas Poell

(2)

University of Amsterdam Faculty of Humanities

Graduate School of Humanities Research Master Media Studies

Master’s thesis

Gephi and Its Context: Three Oppositions in the Epistemic Culture of Network Analysis and Visualisation

Supervisor:dr. Bernhard Rieder Second reader:dr. Thomas Poell Date of completion:29/06/2018

Cover image: A fragment of search results with a query ‘Gephi’ on Google Images search engine, retrieved on 28/06/2018

Acknowledgments

A number of people have been very important for this thesis to become a reality. First of all, of course, I would like to thank my supervisor dr. Bernhard Rieder for not only his expertise, inspiring conversations about the work, but also an attentive ear, positive feedback and good laughs at the right moments. I would also like to offer a thank you to my second reader dr. Thomas Poell for reading and evaluating this thesis, not to forget wonderfully coordinating our study programme. Very importantly, this thesis would have been impossible without the interviewees — a huge thank you to them for finding time in their busy schedules. Especially, I would like to thank Mathieu Jacomy, for not only talking with me, but also for creating Gephi. I am particularly grateful to family: my caring parents for both emotional and financial support, my brother Darius and Dominyka — for also often taking care of my physical wellbeing. Finally, a warm thank you goes to my friends for the support-group-chats online and making the long hours, in and out of the library, fun. Cat, Pijus, Georgina and Pija have also been very helpful with proofreading parts of the thesis.

(3)

Abstract

This thesis is interested in the interrelation of knowledge production with the use of research tools in humanities and social sciences. More explicitly, it looks into the practices of employing Gephi software for network analysis and visualisation in new media studies. The thesis starts by providing a literature context of relevant studies. It is then followed by the basic theoretical ideas from laboratory as well as science and technology studies. Subsequently, an overview of network analysis and visualisation is presented through its historical changes. An inductive empirical analysis was conducted, which consisted of 7 semi-structured interviews with researchers in the field and one of tool’s creators and an analysis of 17 published academic papers. The findings reveal and discuss three oppositions in the relationship of Gephi and its context: tool vs. method, black-boxing vs. unboxing and exploratory vs. explanatory purpose of use. The thesis concludes not only with an overview of the findings and general observations concerning the epistemic culture in the research area, but also with recommendations for the field of study and the developers of the tool.

Keywords: research tools and methods, network analysis and visualisation, Gephi, research practices, ethnographic approach

(5)

Introduction

The presence of computers in the western post-industrial society is undeniable. Humans have been accompanied with everything smart — from phones to watches, trackers, fridges or homes. This ubiquitousness of the digital in people’s everyday lives has come to provide not only private companies but also academia with the resulting digital traces, or, in other words, Big Data. The advances in computational technologies, furthermore, have brought about both new hardware and software to the research environments. Together, these developments have produced a shift in how research is conducted in humanities and social sciences, and, more narrowly, new media studies. Otherwise known as the computational turn1(Berry), it has been related to an increased use of numerous computational methods and tools for analysis in various disciplines, including humanities and social sciences2. In turn, these overall changes have not only brought new possibilities and struggles that existing fields (e.g. sociology (Savage and Burrows; Uprichard et al. etc.) have to respond to, but also introduced whole new research areas, such as digital humanities3 and digital methods4.

Arguing that academia should be reflexive towards its practices, various authors have since addressed this rise in digital approaches for studying both online and offline. Among the many deliberations, the field of new media has been concerned with the opposition of natively digital vs. digitised methods (Rogers, The End of the Virtual) or studying the medium as opposed to studying the social (Venturini et al.; Marres and Weltevrede; Rogers, Digital Methods). One of the important lines of critical argumentation in relation to post-computational-turn changes is concerned with the use of various research tools and the effects of it to the larger field. Some of the most renowned critique in the humanities focuses on the way the field is becoming dominated by the use of tools and how this forces a turn away from theory. For example, Alexander R. Galloway states that ‘[t]hose who were formerly scholars or experts in a certain area are now recast as mere

1_{The ‘use [of] technologies to shift the critical ground of their concepts and theories’ (Berry 11) in arts, humanities and social}

sciences, as well as numerous other disciplines

2_{Examples of research tools: data gathering and preparation (e.g. Netvizz, DMI-TCAT, Table 2 Net, Googlescraper, Image Scraper,} Instagram Hashtag Explorer etc.), analysis (Gephi, CorText, NLTK), visualisation (e.g. Gephi, Raw Graphs etc.)

3_{See: Kirschenbaum}

(6)

tool users beholden to the affordances of the tool — while students spend ever more time mastering menus and buttons, becoming literate in a digital device rather than a literary corpus’ (‘The Cybernetic Hypothesis’ 127).

A large area of tool-critique, however, is concerned with, more explicitly, the relationship between different specifics of tools, methods and the larger research field. Research by Emma Uprichard et al., for example, analysed how the developments and transformations of the statistical analysis tool SPSS5 have contributed to the changes in sociological knowledge over time and demonstrated the impact of various computing technologies on the larger field. As one of the particularities of research tools, the literature has addressed the graphical user interface (GUI) and the variety of problems that come with it. Such critique would argue, for example, that interfaces play a role in how tools ‘shield us from the complexity of thought’ (Tenen) and that in order to use them for research one needs to understand them better than the tool makers (Tenen). Likewise, Bernhard Rieder and Theo Röhlehave argued that the interface not only black-boxes the code, but it also makes the overall research method unclear (‘D.M. From Challenges’). This thesis is in line with the above-discussed problematics, as it is interested in the relationship between research tools and methods as well as the larger academic field. However, the previously discussed critique often approaches the post-computational-turn humanities and social sciences from a quite broad point of view — often looking at some overarching problematics. The current research, by contrast, is concerned with one specific, but widespread, research method, that is, network analysis and visualisation.

Defined in a very straightforward manner, social network analysis can be described as a ‘strategy for investigating social structures’ (Otte and Rousseau 441). It looks at the relationships between actors as opposed to simply the properties of those actors alone. While networks do not have a particular form, they can be visualised in a number of ways. The most common in our field employs an imagery of the actors as nodes and the relationships between them as edges, organised according to a specific layout. Such network visualisations have been critiqued often and from different perspectives. For example, Galloway has famously criticised the current tendency of everything being visualised as a network and the response to these networks being more networks (Network Pessimism). By others these visualisations have been referred to as

(7)

‘hairballs’, (Rochat; Krzywinski et al.) that fail to represent anything or are too difficult to grasp, and critiqued for ‘inherent unpredictability, inconsistency and lack of perceptual uniformity’ (Krzywinski et al. 628). Yet network analysis and visualisation still appears to be a mid or even end-point of a vast number of studies (e.g. Adamic and Glance; O’Callaghan et al.; Larsson and Moe; Herdağdelen et al.) and does not seem to be vacating the field now or in the near future. The popularity of the method, as well as its long history stemming from a complex interrelation of different scholarly fields, already makes the method worth the inquiry. The momentousness of this investigation is also amplified by the abundance of research tools, developed specifically for the purpose of network analysis (e.g. Tulip, Cytoscape, Pajek, Socialaction, GUESS, Citespace, SCI2, Gephi etc.).

One of the research tools used for network analysis and visualisation in media studies stands out from the rest. As a tool in practice used for both analysis (various calculations and metrics) as well as visualisation (producing network imagery using different layout algorithms) of networks, Gephi is a particularly interesting and noteworthy example. Moreover, it is highly interwoven with other software, as it is an end-point with regards to their output files. Furthermore, Gephi has been widely recognised not only in the humanities and social sciences, but also across different fields (such as natural sciences and mathematics). The overall citation count of the 3-page-long paper by Bastian et al. that introduces the tool is currently well over 3000 and counting. This, however, excludes the non-academic uses of the tool such as in the case of journalistic investigations. Some aspects that could be contributing to the wide utilisation of this tool in new media studies is that Gephi is open source and has a graphical user interface. This makes it easy to use for a scholar who is not trained in computer science, mathematics, programming or other relative fields. Since that is often the case in humanities and social sciences, the provided detailed graphical user interface thus makes a variety of research possible in the first place.

Taking into account the previously introduced critique of graphical user interface, Gephi is a rather distinct example. In their 2009 paper presenting the software Gephi creators state that ‘[a]s well as being technically accurate and visually attractive, network exploration tools must head toward real-time visualisations and analysis to improve the user’s exploratory process’ (Bastian et al. 361). Throughout their paper, in general, a great amount of emphasis is placed on a better user-researcher experience. In other words, the creators of

(8)

this software seem to have put effort into not only making the tool easy to use, but also, the method behind it — more transparent. This raises further interest into the graphical user interface of Gephi.

To this day, however, relatively little scrutinisation has addressed the tool at the focus of this analysis. The pieces of critique introduced earlier serve as informative examples, but they are largely approached from the area of digital humanities and mainly text mining methods and tools (such as NLTK6) (e.g. Tenen). Moreover, even fewer studies have engaged with Gephi from a perspective that questions the academic tradition of the use of this tool and its possible implications to the larger field. More explicitly in relation to Gephi, Rieder and Röhle address it as one of the examples in their account of digital methods challenges. They state that the problem of Gephi’s use comes from the fact that it has made ‘network analysis accessible to broad audiences that happily produce network diagrams without having acquired robust understanding of the concepts and techniques the software mobilizes’ (‘D.M. Five Challenges’ 118). Another study that has paid significant attention to this software addresses the question of accountability in relation to the researchers’ use of Gephi. Wieringa et al. define it in terms of being both ‘open to inspection’ (4) and ‘being competent in assessing the subject matter’ (5). In other words, they address the use of Gephi from the perspective of research documentation. They also present a plugin for the tool and discuss a possibility for a ‘reflexive approach to one’s research practice’ (13) by taking field notes. These two studies reflect the majority of the inquiries into the academic tradition of the use of Gephi.

In addition to the previously presented research, however, some academics on a broader scale of the field have called both for a critique of not only tools, but also their users (Rieder and Röhle, ‘D.M. From Challenges’) in social sciences and humanities as well as for engaging more with a structured methodology and inquiring into methodology itself (Clement). Christine Borgman argued in 2009, that ethnographic and other social studies should be implemented to study digital humanities. Likewise, Tanya E. Clement, in ‘Where Is Methodology in Digital Humanities?’ states that a social analysis of the use of digital methods in digital humanities is needed. While some of this research addresses the topic in a different field, one can argue that the case is similar in media studies - a field largely embedded in the humanities employs methods that come from more classical scientific fields. Authors in new media studies (e.g. Wieringa et al.) have also

(9)

called for an ethnographic research in the field. All in all, an investigation into the methodology, the user perspective and an ethnographic approach to the use of tools has been called upon, yet largely unrealised. This thesis aims to reflect on the forming research tradition in post-computational-turn field of new media studies. As a study interested in the interrelation of digital research tools, methods and their relationship to the larger discipline, this written work focuses on a case of a specific piece of software. The Gephi network analysis and visualisation tool was chosen due to its unique situation in new media studies. As a sophisticated and widely-used software that aims to make network analysis and visualisation more accessible to its user, Gephi is important to scrutinise. However, as has been observed, there is an overall lack of an overarching understanding of how this tool is used, what the academic tradition it is embedded in is, and how it relates to the software itself. Having in mind that computational approaches to research are not likely to decline in humanities and social sciences, there is a need to critically engage with them. Furthermore, if one were to follow an approach rooted in science and technology studies, an absence of empirical ethnographic research can be observed in relation to not only the specific case of Gephi, but the larger field in general. By problematising the researchers’ engagement with the tool, the epistemic community can begin to grasp how it should understand the knowledge that is produced in such apparatus and adjust the practices of using the tools, both on an individual and field level. In trying to tackle these varying problematics in the field, this research aims to study the network analysis and visualisation practices with Gephi. How do researchers engage with this tool, make and justify methodological decisions, document the research process and communicate it? What observations can be made about the larger epistemic tradition and its interrelation with the instrument? How can the use of Gephi in the research field be problematised and what recommendations can be provided for seeking beneficial changes?

This thesis is divided into three chapters that will in succession present, first, the context of this study in relevant literature; second, the undertaken methodological approach and, third, the findings and the discussion of the research. The first chapter will look at the main themes in preceding research, discuss the theoretical background of ethnographic laboratory studies as well as present the main aspects of the historical complexities of network analysis and visualisation. The second chapter will focus on the empirical methodology of the current study, presenting the mixed-methods approach, consisting of semi-structured interviews with researchers, as well as exploring how Gephi use is communicated in academic papers. The

(10)

third chapter will present the findings and discuss them via the main themes that stem from the conveyed inductive approach analysis. Lastly, the conclusions will not only summarise the findings, but also provide recommendations to the field.

(11)

Chapter 1: Context in relevant literature

It has already been briefly mentioned in the introduction that the relationships of tool-use for research and the overall knowledge construction process has to some extent been addressed in previous studies. Pooled together from disparate disciplines, this literature context is relevant to delve into before moving on to the specificities of the current research. To begin the overview, I would like to address the larger milieu of studies that have in one way or another touched upon the problematisation of computer use with regards to knowledge production. As early as the start of an academic journal Computers and the Humanities in 1966, attention was turning to how computational-tool-use differs among fields with regards to the nature of their study objects7. Lately, the topic has been addressed by digital humanities scholars, who have endlessly questioned the very definition of their field (e.g. Schreibman et al.; Kirschenbaum; McCarty; Svensson). Moreover, digital sociology has inquired into both how one can sociologically study the digital as well as how digital methods can be used for it (e.g. Marres). The area of Computer-Supported Cooperative Work (CSCW)8 has looked at how computers can be used to support work situations, including collaborative research work (e.g. Kouzes et al. ). Finally, the use of digital methods has been problematised from the perspective of the ‘lack of separation between medium and message’ (Venturini et al. 1). In awareness of these and other miscellaneous accounts, the current research occupies only a small and very specific segment that lies at an intersection of some of these and certain other fields.

The following chapter will thus provide a background of the core ideas from the literature, dividing the overview into three parts. First, the chapter will explore the main themes in recent inquiries into tools and methods in humanities and social sciences that this thesis was prompted by. In the second section an account on laboratory studies and the related notions of epistemic culture, laboratory, inscription device and context will be introduced. The last section will briefly explore the historically complex method of network analysis and visualisation.

7_{See: e.g. Lieb} 8_{See: Greif}

(12)

1.1: Previous inquiries into computational tools and methods

in humanities and social sciences

Beginning the overview of some core themes in the literature that deals with the use of computational technologies and methods in humanities and social sciences, it is noteworthy that while some studies have paid attention to the object of the current analysis from a computational perspective (e.g. Yang et al.), very few inquiries have problematised Gephi from the point of view of social sciences and humanities. A distinguished text by Rieder and Röhle (‘D.M. From Challenges’) exemplifies this tool when making broader claims about the state of the field. The authors use it as one of the three cases of research techniques for consideration and argue that the current use of the tool ‘leads to a lack of awareness of the layers of mediation network analysis implies and thus to limited or essentialist readings of the produced outputs that miss its artificial, analytical character’ (118). In other words, these authors critique the current use of network analysis that — through tools like Gephi — produces and analyses outputs without taking into account the complex methodological considerations involved in their creation. They further provide a number of suggestions about how a better understanding of the method behind Gephi can be acquired. While there are more studies concerned with Gephi software to varying degrees, the following section will not be limited by this criterion. It will rather aim to distil a number of themes around the inquiries in the broader field, that have thus far been concerned with research tools or methods.

Tools and Methods.

Various authors in the literature, concerned with the post-computational-turn humanities and social sciences, allude to the question of the interrelation between tools and methods. In more specific cases, the use of one as opposed to the other. From the Greek word methodos — ‘a pursuit of knowledge’ (Liddell et al.) — a research method is a systematic approach to acquiring a grasp of a specific object or phenomenon. A tool, then, can be considered to be an instrument that is needed to implement (a) method(s) — it can have a multitude of forms, from standardised psychological tests to programming languages and graphical-user-interface-supported software. In others words, methods are ‘what remains long after the tools pass into obsolescence’ (Tenen). The scholars state, for example, that ‘[i]f students and researchers are trained in using […] tools without considerable attention being paid to the conceptual spaces they mobilize, the outcomes can be highly problematic’ (Rieder and Röhle, ‘D.M. From Challenges’ 117). In other words, they problematise the practice of learning to use tools as opposed to understanding and applying the

(13)

methods, that stand behind them. While it is unrealistic for academics to promote a complete disengagement with computational tools, they do seek to raise awareness of the understanding of how these tools operate. It is, however, increasingly difficult to make a division between tools and methods in the age of the digital. Keeping in mind the distinction between natively digital and the digitized (Rogers, The End of the Virtual) tools, it can be argued that there might be tools and methods that only exist in conjunction. To make the cases more clear, one may look at statistics as opposed to artificial-intelligence-driven research methods. Most of the statistical calculations, provided by software — like previously mentioned SPSS — can be also calculated manually, as has been prior to the digitised technologies. What is problematised in relation to them is, to a large extent, not the tools themselves, but the methods behind them9. On the other hand, in the complex cases of artificial intelligence used for research such as in natural language processing, the question of the tool-method boundary grows ambiguous as it becomes difficult to discern the two — machine learning cannot occur pre-machine.

Some authors thus suggest it is not necessary to get engaged with the method in its pre-computational form (if that exists) - it is possible and sufficient to learn (about) the method by inspecting the tool. In the case of Natural Language Toolkit (NLTK), Dennis Tenen argues that ‘[u]nlike some other tools of its kind, NLTK is particularly good at revealing its methods. Its codebase is open to inspection; it is easy to read; and it contains much commentary along with links to related research’. Such theorising is, however, problematic, since different tools can manifest their methods to varying degrees, while still the ‘[t]ool can only serve as a vehicle for methodology’ (Tenen) as opposed to providing a completely direct view on how it operates. The notion of black-boxing, which will be discussed next, is highly relevant here.

Black-boxing.

One of the most prevalent themes within the relevant literature problematises research tools as black-boxing devices. To cite Bruno Latour:

“The word black box is used by cyberneticians whenever a piece of machinery or a set of commands is too complex […]. That is, no matter how controversial their history, how complex their inner workings, how large the commercial or academic networks that hold them in place, only their input and output count” (2–3).

9_{for example, the famous debates about the (mis)use of the probability value (p-value) (see Wasserstein and Lazar; Ranstam;}

(14)

In other words, black-boxing refers to keeping parts of the process in the dark because they are regarded irrelevant to show, too difficult to understand or kept a secret for other reasons. More specifically, in the case of digital objects (also, research tools), what is black-boxed, often refers to the code that runs behind the graphical user interface.

Just how tools and methods vary from one case to another, their black-boxing may also have dissimilarities. Bernhard Rieder and Theo Röhle, for example, layout the process of black-boxing on three levels. First they list the ‘practical possibility to access the most obvious layer of functional specification, a tool’s source code’ (‘D.M. Five Challenges’ 76). Second, they discuss ‘code literacy’ (76), and last they present the opaqueness of methods (very often machine learning) that exist only in their digital equivalents, thus may never be fully understood by people. According to these authors, black-boxing refers not only to the possibility to read the code, but also the capability of its user to open and comprehend the research method implemented.

In relation to the three stipulated levels, black-boxing can be perceived as hindering ‘our ability to understand the method, to see how it works, which assumptions it is built on, to reproduce and criticise it.’ (Rieder and Röhle, ‘D.M. From Challenges’ 112). To better understand the problematics of black-boxed research instruments, a quote from Emma Uprichard et al. is relevant. They state that in the case of SPSS, the black-boxedness of the software transformed the user into ‘increasingly dependent on the software and computer technology to think statistically and to perform analysis’ (612). Axel Bruns similarly argues, that:

There is a substantial danger that social media analytics services and tools are treated by researchers as unproblematic black boxes which convert data into information at the click of a button, and that subsequent scholarly interpretation and discussion build on the results of the black box process without questioning its inner workings

That is to say, some of the risks of using black-boxed instruments stem from becoming reliant on the software to be able to use a specific research method as well as being unable to judge the inner functioning of the research tools and, consequently, interpret the viability of the results that the software provides one with. A major topic within discussing black-boxing of research tools deals with the question of how they may be unboxed. While arguing that when using tools we should ‘master them from the inside out’ (Tenen), authors also recognise that it is complicated taken into consideration how many layers of black-boxing they are

(15)

wrapped in. Through the example of NLTK, Tenen shows how graphical user interfaces may remove the user from the implicit logic of the method - ‘each level of abstraction in the movement from statistical methods, to Python code, to graphical user interface introduces its own set of assumptions, compromises, and complications’. Tenen further argues that the inner workings of complex research mechanisms ‘lie beyond the full comprehension of the casual user’. However, even though a complete understanding of the functioning of the tool requires a significant level of work and sometimes is even impossible, according to Tenen investing into understanding what is hidden is a major part of the research process.

By contrast, some authors provide a different view towards dealing with black-boxes. For example, Knorr Cetina (Epistemic Cultures) argues that some boxes in the research process should remain unopened(Epistemic Cultures). Similarly, Benjamin M. Schmidt states that ‘[p]ast a certain point, humanists certainly do not need to understand the algorithms that produce results they use; given the complexity of modern software, it is unlikely that they could’ — tying back to the example used when discussing the blurring line between tools and methods. Probably one of the most prominent recent interrogation into the question of black-boxing is provided by Johannes Paßmann and Asher Boersma in their chapter of The Datafied Society entitled 'Unknowing Algorithms: On Transparency of Unopenable Black Boxes’. The authors separate the notion of transparency into formalised and practical. As formalised transparency they define practice that ‘largely tries to obtain a ‘more positive knowledge’ on ‘the content of a black box’ (140), whereas practical transparency ‘does not try to open black boxes, but to develop skills without raising the issue of openability’ (140). In other words, these authors argue that black-boxes that ‘remain intransparent in a conventional, formalized sense [are] in no way problematic for the scientist in the practice of research’ (141), since what matters is learning to work with the ‘known unknowns’ (145). That is, to be able to say which parts of the black-box output can(not) be trusted. All in all, the discussed authors argue that not every black box can or needs to be opened, while also Paßmann and Boersma suggest learning to deal with what ‘remains black after all’ (140). To conclude, the discussed authors, on the one hand, disagree about some fundamental ideas, such as how realistic it is to imagine humanities scholars developing sophisticated research tools themselves10, or how and if black boxes of research tools should be opened. On the other hand, they do seem to agree that the role of

(16)

the user should not be underestimated when discussing the black-boxing effects of software. For example, Rieder and Röhle state that ‘[a] critique of digital tools is incomplete without a critique of their users and the wider settings they are embedded in’ (‘D.M. From Challenges’ 116). Dennis Tenen, similarly, criticises a concept he refers to as the ‘lazy mode of thinking’, which touches upon the interpretative part of research — for example, cases when unsupervised machine learning algorithm returns an output that needs to be explained by the researcher. He states that ‘[t]o use such tools well, we must, in some real sense, understand them better than the tool makers. At the very least, we should know them well enough to comprehend their biases and limitations’. To summarise their approaches, the presented studies raise a question of how interrelated black-boxing is with the involvement of the user and the larger academic tradition. This thesis will, consequently, aim to further investigate this link.

Research Documentation and Communication.

The final theme that has been touched upon in previous research is concerned with how digital-tool-based empirical new media studies are communicated in scholarly publications. To be more particular, they look at how such publications present the methodological steps taken and the findings they arrived to as well as inquire into the possible roots and implications of that. Wieringa et al., for example, provide an overview on how the use of Gephi is introduced in scholarly articles. They look at a number of publications of research that implement this tool and present an account of what methodological details are presented. The authors argue that two types of actors are at interplay in the knowledge construction through the use of Gephi — ‘human and non-human’ (3), or the user and the tool. Based on the notion of ‘situated knowledge’11, they argue that in order to expose the process of the knowledge construction, certain accountability is needed. In their opinion it can be based on the documentation and presentation of one’s research process, for which the authors suggest a method of field notes and introduce a Gephi plugin. In other words, the research documentation should take into account the idea that researchers are not neutral observers of reality, thus it is important to present how specific knowledge was constructed through the methodological notes of the research process.

On the other hand, it has also been expressed in the literature that clear and detailed presentation of one’s methodological steps is important for the replicability of research results. For example, Axel Bruns notices

(17)

that currently very few research studies ‘insert any substantive discussion of the benefits or limitations of the particular Gephi network visualisation algorithms they have chosen, or of the specific visualisation settings which were used to direct the algorithm itself’ and sees it problematic for the possibilities of research reproducibility. However, he also looks at the publishing media and presents a number of its possible limitations that may cause such a trend in humanities and social sciences, naming spatial, temporal and format limitations, which may restrict the scholars to invest more focus into the tools they are using in the publications. All in all, the research documentation and presentation, while important for the reproducibility of results, is related not simply to the decision by the researchers, but also the standards dictated by the academic environment — in this case, the academic journals.

All in all, the scholars concerned with the documentation and communication of research have highlighted the questions of accountability and reproducibility of results. They have not only highlighted the need to present the methodological steps accurately and thoroughly for the above-mentioned reasons, but also have touched upon the academic environment, which is in different ways relevant for the documentation of the research process. While it is important to account for ‘a particular framework: a paradigm, a (socio-economic) background, a discipline, and so forth.’ (Wieringa et al. 4) one’s ideas are embedded in, the specific research setting (for example the publishing medium) also may be restricting the possibilities of doing so accurately and rigorously.

________

This first section of the chapter has shown that studies have problematised the use of tools in the humanities and to some extent discussed the case of Gephi. However, a few aspects should be taken into consideration. First of all, many of these studies have been conducted in a close, yet different field of digital humanities. This should not be seen as highly problematic since the fields are intertwined and many of the discussion points can be transferred. However, one may remark that in a field that has a number of quite central, highly used research tools, research should be conducted that scrutinises the specific cases. Furthermore, the majority of these studies approach the object from a largely theoretical point of view and tend to provide arguments about the research tradition without introducing empirical indications. Only one study — by Wieringa et al. — was detected that provides such findings. However, neither this, nor any other research in our field aimed to look at the questions of the tool-use inductively, that is, give voice to the subject and take

(18)

into account questions and problems that stem from the research practices in actuality. Even though the discussed studies raise the question of user scrutinisation, an ethnographic research of laboratory studies tradition has not been conducted in the humanities and social sciences, more importantly, this approach has not been taken to study Gephi. The following section will introduce the original ideas, stemming from this approach.

1.2: New media studies as a laboratory: epistemic culture,

inscription device and context

The current thesis aims to position the research environment, that Gephi use is embedded in, as a laboratory. While it may not seem as an obvious choice to discuss a humanities or social sciences discipline in terms of a concept more often found in classical scientific disciplines, Knorr Cetina’s understanding of it seems more than relevant. The author presents her account of this notion by opposing it to the understanding of laboratory as a simply physical location (Knorr Cetina, Epistemic Cultures). Rather, the essence of a laboratory to her is ‘based upon the premise that objects are not fixed entities that have to be taken "as they are" or left by themselves’ (Epistemic Cultures 27) and that they ‘allow natural processes to be "brought home'' and to be made subject only to the conditions of the local social order’ (Epistemic Cultures 28). That is, instead of working with objects that occur in the ‘nature’, laboratory is the entity that allows one to work with their ‘object images or with their visual, auditory, or electrical traces’ (Epistemic Cultures 27). One can thus say that a laboratory is defined rather not by a physical location, but by the activity of working with produced extractions of objects.

In the case of media studies, then, the research environment may be quite difficult to envision as a physical locality. Similarly to other studies (e.g. Kouzes et al.; Beaulieu) that have addressed the question of more distributed laboratories, the one that is discussed in this thesis is rather an open space, since it is rare that the academics work in close physical proximity with one another. Moreover, as argued by Knorr Cetina, ’[n]ow consider the laboratory in social science areas. It does not, as a rule, involve a richly elaborated space — a place densely stacked with instruments and materials and populated by researchers.’ (Epistemic Cultures 35). It does, however, on the larger scale, involve other researchers, teachers and students, software and hardware

(19)

that is used, research documentation, both read by scholars as well as produced in their laboratory. Most importantly, it involves electrical traces of varying phenomena, such as users’ activity online — in other words, cooked12 data (Bowker). Explained differently, while the laboratory in humanities may not seem similar to the one from ‘hard’ sciences, it can be understood as corresponding to Knorr Cetina’s understanding of the notion.

In order to look more closely at the laboratory in humanities and social sciences, one may turn to laboratory studies — a research area that stems from Latour and Woolgar’s ethnographic account of research practices in science laboratories, covered in the book Laboratory Life. It is important to note that the field of science, technology and society studies (STS) can be seen as having originated from laboratory studies. The fundamental argument of STS is ‘that scientific knowledge and technologies do not evolve in a vacuum’ (Law 12). That is to say, a number of aspects of the environment are significant with regards to the construction of knowledge. In fact, John Law also notes that ‘ethnographers of science are usually more or less

constructivist. That is, they argue that scientific knowledge is constructed in scientific practices’ (19). In STS terms, then, knowledge is not something that seeks to explain the ‘real’; rather it is a result of a quite messy research process and its context. It seems very important then to look not only at the research practices themselves, but also deconstruct elements from their environment in order to see their effect on the knowledge production in the field. Taking inspiration from the laboratory studies (mostly work by Latour and Woolgar and Knorr Cetina), the remaining part of this section will introduce the concepts from this field that are most relevant for the current thesis, that is inscription device, instrument and its context as well as epistemic culture.

Let us start by looking at the notion of epistemic culture with regards to knowledge production processes. Knorr Cetina explains that ‘[e]pistemic cultures are cultures that create and warrant knowledge, and the premier knowledge institution throughout the world is, still, science.’ (Epistemic Cultures 26). In her words, the notion epistemic culture ‘disrupts the idea of the epistemic unity of the sciences and suggests that the sciences are in fact differentiated into cultures of knowledge that are characteristic of scientific fields or research areas, each reflecting a diverse array of practices and preferences coexisting under the blanket

(20)

notion of science.’ (Knorr Cetina and Reichmann 873). To put it briefly, the author’s account of epistemic cultures refers to the idea that specific research fields and areas build up their own self-referential cultures with certain methodological practices. It is thus important to analyse them separately; as ethnographic accounts from one epistemic culture, cannot be easily transferred to another. With regards to the idea of laboratory, discussed earlier, an epistemic culture can then be understood to manifest through the social interactions in the laboratory.

To think back about the extractions or traces of objects referred by Knorr Cetina to constitute a laboratory, one can also approach them from a different perspective — seeing them as inscriptions produced through the so-called inscription devices. Latour and Woolgar introduce the notion of an inscription device originally by defining it as ‘any item of apparatus or particular configuration of such items which can transform a material substance into a figure or diagram which is directly usable’ (51). According to the same authors, the way that an apparatus is arranged may be of significant importance for the production of a useful inscription. That is, the properties or settings of the device, the way the researcher uses them may be responsible for what kind of knowledge they present us with.

In research practices, accordingly, one can find instruments that come in contact with materials analysed and return an output that the researcher can grasp - such as mathematical calculations or visualisations. According to Latour, ‘an instrument [is] any set‐up, no matter what its size, nature and cost, that provides a visual display of any sort in a scientific text.’ (68). Gephi, for example, can be considered a research instrument. To clearly separate the notions of an inscription device and research instrument, it needs to be noted that any research tool that produces an output to be used in the analysis can be called an inscription device. It is, however, not analogous the other way around — not every inscription device is an instrument. That is to say, Gephi can be regarded both an inscription device, because it produces inscriptions, as well as an instrument. On the other hand, there may be other items in the research apparatus that produce inscriptions in an academic setting (for example, researchers’ senses and perception can be considered an inscription device) as well as there may be different layers of inscription in one instrument.

Latour and Woolgar also theorise that a specific context is relevant, together with an inscription device, (or, an instrument), for the knowledge construction process. One may regard a number of different aspects to be

(21)

the context - it can be social or economic factors on a larger scale, or the field within which the inscription device is utilised, the specific social relationships in the laboratory setting, and other facets of the particular research environment. From Knorr Cetina’s perspective, a specific epistemic culture may be considered to be one part of the context that the research process is surrounded by. As some of the contextual aspects of an epistemic culture this author presents the discrepancies between the ways that research is conducted and how it is presented in papers. This can be seen as one aspect of the context of a specific inscription device. Another example could be seen to be the learning process — how one enters a specific research context, how they gain an understanding of that specific research instrument. On a larger scale, the context of an inscription device can consist of many aspects, however, this research will take the mentioned examples as some of the main ones.

To give an example of a more recent study, which employs the notion of inscription device to analyse research practices in sociology, an article by Emma Uprichard et al. is of relevance. The authors consider the statistical analysis software SPSS as an inscription device and look at how the change in technology of statistical analysis techniques has influenced the sociological form of knowledge. In a substantially historical account they trace the changes in the instrument together with the adaptations in the field (the context). Uprichard et al. attribute a number of turns of the latter to the arrival of the former, such as a shift in pedagogy: ‘[t]extbooks and student guides shifted to having much less discussion on the processes and theories involved in interpreting and constructing quantitative representations of the social world than they had done before the widespread use of the software.’ (611). Overall, the transformed pedagogy was seen as a process that largely taught how to use the tool rather than how to perform statistics. This research by Uprichard et al., overall, can be seen as an inspiration for the current study.

This thesis is concerned predominantly with a specific instrument — the network analysis and visualisation software Gephi — as well the larger environment it is embedded in. It will inquire into the laboratory of Gephi users, the context of this tool and the epistemic culture that surrounds them. Some of the contextual aspects, that this thesis will be interested in, are the learning process of applying network analysis through Gephi, the decision-making when using the tool, the documentation of the methodology and communication of the findings to the field etc. Moreover, this thesis is interested in how the inscription device and the context shape each other. As part of the context of Gephi, one can also consider the method of network

(22)

analysis and visualisation together with its historical complexities — it will be discussed in the upcoming section.

1.3: The method of network analysis and visualisation

As argued by Rieder and Röhle, to be able to critically engage with a tool, it is key to be attentive ‘not just to the software form, but to the actual concepts and methods expressed and made operational through computational procedures’ (‘D.M. From Challenges’ 117). In other words, a crucial component of getting a grasp of the context of Gephi is acquiring a pertinent understanding of the network analysis and visualisation method. This thesis argues, however, that in the case of this specific method it is not enough to look solely at the contemporary understanding of it — due to its complex and difficult history, one needs to take into account the aspects of the method that occurred in the past. Furthermore, a look at network analysis and visualisation in general would allow for both a better understanding of the context of the instrument in the Latourian sense, as well as provide a better grasp of the mechanisms at play in the software. The following section, therefore, will address the main aspects of the method together with some historical characteristics that may be seen as contributing to the current understanding and use of network analysis and visualisation. It is significant to note in the beginning of this overview that network analysis and visualisation is embedded in a very wide and historically complex field, consisting of psychology and sociology schools, stemming from cognitive Gestalt psychology (work by Kurt Lewin13 and Jacob L. Moreno14) and structural-functional anthropology15 as well as graph theory16 in mathematics. However, the scope of this research is neither vast enough to introduce the overall field in great detail, nor is it relevant for its focus. Therefore, I would like to draw the reader’s attention to three relevant facets of network analysis and visualisation. First — the origination of the method and its underlying ideas; second — visualisation in network analysis; third — a brief account on software that implemented network analysis computationally.

13_{See: Lewin, Principles of Topological Psychology; Lewin, ‘Defining the “Field at a given Time.”’}

14_{See more: Moreno, Who Shall Survive?; Moreno, Sociometry and the Science of Man.; The Sociometry Reader} 15_{See: Lévi-Strauss}

(23)

When discussing the development of the method of network analysis and visualisation, the first aspect that needs to be addressed from a historical point of view is the interrelation of social sciences and graph theory in mathematics. In his article considering the PageRank algorithm, Bernhard Rieder provides an overview of the history from exactly this perspective — he states that ‘[r]ather than being imported into these disciplines as a ready-made mathematical toolkit, graph theory was shaped in no small part by the measuring fever in the field of ‘small-group’ research, founded by Kurt Lewin and Jacob L. Moreno’. The mentioned authors as well as the mathematician Frank Harary could be considered to be some of the names that contributed most to the early developments of network analysis and visualisation.

The foundations of the method can be marked by sociometric tests in Moreno’s psychological work in ‘small-group’ analysis, presented in the book ‘Who Shall Survive?’. Moreno presents his method by explaining that social ties between group members may show the ‘amount of organisation’ (11) that the group has. His method was based on observations and interviews, but also, and more importantly, sociometric tests. In his book he explains that ‘[t]he sociometric test requires an individual to choose his associates for any group of which he is or might become a member’ (11). Based on this ‘attraction’ and ‘repulsion’ between actors, the first analyses were conducted, which used very simple ‘point and line’ diagrams drawn by hand (for an example see the left image in fig. 1, in the next page), which could be considered very distant predecessors of the current computerised network visualisation examples.

In terms of the modern social network analysis, as an organised paradigm for research, Linton Freeman introduces it in relation to four features — it is ‘motivated by a structural intuition based on ties linking social actors; […] grounded in a systematic empirical data; […] draws heavily on graphic imagery, and […] relies on the use of mathematical and/or computational data’ (3). One of the mentioned aspects, however, was not met at the early development moments of Moreno’s work - that is, they ‘included no mathematical or computational models at all’ (Freeman 38). In fact, Moreno’s method was criticised for the lack of actual mathematic method, but in return generated a lot of interest from professionals with a background in mathematics (for example, by Forsyth and Katz, Paul Lazarsfeld17), which led to their contributions (Rieder;

(24)

Freeman). This began a long history of the two interwoven fields with input from, among others, Frank Harary and Kurt Lewin.

Figure 1. A sociogram and a matrix visualisation of the same sociometric data18

With regards to the visualisation aspect in network analysis, also listed above by Freeman, the introduction of more mathematical techniques also brought in an update. After Moreno’s original ‘point and line’ diagrams, Forsyth and Katz introduced the first quasi-algorithmic technique for visualising networks. It was a matrix, supposed to provide a less confusing and more objective overview (see the right image in fig. 1, above). The introduced technique of manipulating the matrix ‘in a way that subgroups would become visible, simply by reordering rows and columns’ (Rieder) provided a more standardised approach to analysis. With time, however, imagery in social network analysis faded into the background of the method, while the more strict mathematical-statistical approaches took the front row. This historical aspect leads well into one of the main points that requires emphasis in this section, that is, that there is no one representation of a network and that graphical imagery is not the central tenet of network analysis. Networks may be visualised in a variety of ways, which is increasingly noticeable in the contemporary uses of various software for network analysis and visualisation. Recently, however, with the rise of network analysis software, the visualisation has again

(25)

become a more focal part of this method. As presented by Krzywinski et al., ‘[i]nvestigators rely on visual interfaces to networks to manage complexity, help create a mind-map of the patterns and cogently communicate their findings’ (627).

Most of the contemporary network visualisations are based on force-directed layout algorithms, which implement physical forces for positioning the nodes. Gephi’s default layout algorithm ForceAtlas2 serves as a good example:

ForceAtlas2 is a force directed layout: it simulates a physical system in order to spatialize a network. Nodes repulse each other like charged particles, while edges attract their nodes, like springs. These forces create a movement that converges to a balanced state (Jacomy et al. 2)19.

While very popular and having different layout versions, the force-directed layout algorithms have also been quite extensively critiqued. Krzywinski et al., for example, point to the reliability of force-directed layout visualisations. They state that ‘the effectiveness of these methods is reduced by inherent unpredictability, inconsistency and lack of perceptual uniformity’ (628). Furthermore, they argue that the unpredictability of such network visualisations is influenced by the fact that they are ‘driven by an aesthetic heuristic that can influence how specific structures are rendered’ (628) and that 'different algorithms generate very different layouts of the same network’ (628)(for an example see the images in Figure 2, in the next page). It needs to be noted, however, that even the same force-directed algorithms may produce different final visualisations of the same dataset (for an example see the middle and right images in fig. 2, in the next page), as they are ‘notoriously brittle: they have many parameters that can be tweaked’ (Munzner 206) and ‘[t]he result varies depending on the initial state (Jacomy et al. 2). Moreover, [t]he process can get stuck in a local minimum’ (Jacomy et al. 2) energy configuration, that is not considered to be the best answer globally. Finally, the network visualisation created with force-directed algorithms is ‘not deterministic, and the coordinates of each point do not reflect any specific variable’ (Jacomy et al. 2). The listed drawbacks emphasise the other difficulties, presented previously in the first section of this chapter, concerning the documentation of the research process. Due to the many parameters and other details that need to be taken into consideration, it becomes overly complicated to document, communicate and reproduce how the network visualisation came

(26)

into being. Consequently, it can be said that force-directed network visualisations have specificities that impede the reproducibility from both documentation as well as design perspectives. However, even though other authors have presented alternatives20, force-directed algorithms remain the most popular approach among social sciences and humanities scholars.

Figure 2. Three networks of the same dataset, but different types of forces. From left to right: Fruchterman-Reingold; ForceAtlas2; ForceAtlas2 with LinLog mode (source: Jacomy et al.)

Among the most widespread network analysis and visualisation tools nowadays, one may find Cytoscape (Shannon et al.), Pajek (Batagelj and Mrvar), GUESS (Adar), Gephi (Bastian et al.) and others. In the field of new media studies, Gephi software is highly favoured for its accessibility as open source software and the user-friendly graphical interface. The tool was first presented by a publication in 2009 introducing it as an ‘open source software for graph and network analysis’ (361). Gephi was from the beginning equipped with a native force-directed algorithm ForceAtlas, which was later updated to ForceAtlas2. Since it is not known how the software came into being and what exactly were the initial ideas of its creators, a lot of questions about the tool cannot be answered. For example, it is unclear how the Gephi software relates to the larger field of network analysis and visualisation historically. Can it be regarded as primarily a research approach that has been packed into a software object? Also, Rieder and Röhle suggest that the humanities scholar implementing this research technique should look at its past (‘D.M. From Challenges’), but it is unclear if the researchers actually see a continuation between the historical and the contemporary uses of the method.

(27)

________

Overall, this chapter focused on three aspects that are relevant for the contextualisation of the current study on network analysis and visualisation software Gephi. First, it introduced the main themes in the literature, which has dealt with this and other research software in the humanities and social sciences. Second, the ideas from laboratory studies were introduced — namely, the notions of epistemic culture, laboratory, inscription device, instrument and its context. Third and last, the method behind Gephi software was presented in relation to its history. The chapter, more generally, criticised the previous accounts of similar research for largely theoretical or anecdotal-evidence-based approaches. It argued that there is a need for inductive empirical research that reflects on the actual research practices and the problematisation that stems directly from them. Moreover, the field of media studies is in need of an in-depth analysis on their native research practices. The following chapter will put forward the methodology of this thesis, which corresponds to the provided arguments.

(28)

Chapter 2: Analysis of Gephi research practices

In light of the introduced theoretical considerations and literature sources, an opening for a particular type of methodological approach can be found in the larger academic corpus. The previous chapter has discussed some largely theoretical considerations with regards to research tools, methods and the research practices, surrounding them. Most of them have been conducted in the field of digital humanities (e.g. Paßmann and Boersma; Tenen; Clement) — an area that is close to media studies, although it often employs different computational methods. Some analyses that have discussed Gephi in particular (e.g. Wieringa et al.; Rieder and Röhle) have not, however, problematised the research practices from an inductive perspective. Finally, tools have been discussed in terms of their connection to the larger research field elsewhere (e.g. Uprichard et al.). This thesis, therefore, aims to fill in the gap of an ethnographic21 bottom-up approach with regards to network analysis and, more distinctly, Gephi software, which is lacking in media studies.

To take a look at some of the reasoning for an ethnographic approach of academic research, John Law argued that ‘[…] ethnography lets us see the relative messiness of practice. It looks behind the official accounts of method (which are often clean and reassuring) to try to understand the often ragged ways in which knowledge is produced in research’ (19). Likewise, a need for ‘ethnographic work in the domain of digital methods and software tools, their use and development’ (11) has been also called for by Wieringa et al. in their previously discussed study on Gephi.

Having taken into consideration and subsequently introduced the need for empirical analysis of research practices in our field, a fundamental characteristic of the current research is its inductivity. The approach taken seeks to provide an emic22 perspective of the subjects in the study — the members of a particular epistemic culture. What is more, such an approach also helps avoid a normatively charged and largely paternalistic narrative with regards to how research tools should be used. This thesis seeks to allow the

21_{Hammersley and Atkinson provide an extended overview of what ethnography can mean in social research today, also defining the}

main characteristics of such an approach, among which: 1) a strong emphasis on exploration rather than hypothesis testing; 2) working with ‘unstructured data’; 3) investigating a small number of cases in depth and from a multitude of perspectives

22_{Defined by Merriam-Webster dictionary as ‘of, relating to, or involving analysis of cultural phenomena from the perspective of one}

(29)

researchers in the field to reflect on their research practices and problematics by, first, engaging in semi-structured interviews with them and one of the tool’s creators. The qualitative mixed-method approach consisted afterwards of an exploration of academic papers in order to triangulate (Miles et al.) the findings and provide an additional data source, having in mind the previously mentioned discrepancies between the research practices and the documentation of them. It is also worth mentioning that the author of this thesis is acquainted with the tool herself, therefore, as a member of the field, is capable of reflecting on its main functionalities as well as the graphical user interface.

Another key element to discuss before moving on to a detailed presentation of the research methods is narrowing the research scope to cases in which Gephi is used for social media research. The reasons for such delimitation are several: first, social media analyses are very prevalent among the field of new media research. Furthermore, in a research field as large as network analysis and visualisation and when analysing the use of a very popular instrument23, there is a need to constrict the analysis to more particular cases. If one were to target the use of Gephi generally, the broader interview questions may be confusing to the subjects while the scope of scholarly paper analysis — unrealistically big for the extent of current research. The chosen delimitation angle, however, is not without grounds — social media24 can be hypothesised to have played a noteworthy role in the recognition of the method itself. It is due to the grammatised friendships, subscriptions, likes etc., that the ‘[s]ocial network sites […] provide rich sources of naturalistic behaviour data’ (boyd and Ellison 220). The gathered data can be viewed as conveniently organised in a way similar to that of Moreno’s sociometrics and 'lend themselves well to analysis through network visualisation’ (boyd and Ellison 220). It may be speculated that with the popularity of such websites and the growing opportunities to gather their data through various tools (e.g. Netvizz25 and DMI T-CAT26), that return a Gephi-supported output, the use of this instrument in the field of new media studies could have increased.

23_{Gephi use is not restricted to humanities and social sciences, therefore there are plenty of cases of it being used in, for instance,}

biology (e.g. Su et al.; Delmont et al.; Bekaert et al.; Xia et al.), medicine and genetics (e.g. Divo et al.; Campo et al.; Zhang et al.) or ecology (e.g. Pocock et al.; De Montis et al.).

24_{The use of a term social networking sites is noteworthy in this discussion — while more prevalent in the earlier days of Web 2.0, the}

term is still used to refer to a part of larger social media, to refer to websites such as Facebook or Twitter, for example. However, the use of this term is increasingly debatable (see, for example, Kwak et al.)

(30)

2.1: Two types of interviews

Stemming from the previously discussed lack of inductive empirical research in relation to the research practices of using Gephi, two types of interviews were conducted. First, practicing researchers were interviewed followed by a conversation with one of the creators of the tool.

The aim of the interviews with the users was to acquire a level of understanding of the research process with Gephi as well as their reflections and perspectives on them — a specific emphasis was placed on the decision-making process in relation to using the tool. Moreover, the interviewees’ considerations with regards to the larger academic tradition were of relevance when organising and conducting the interviews. In general, what was aimed to be understood from these interviews, was the relationship between the justifications of particular methodological choices and the more general understanding of the method and the knowledge of it.

The second type of interview was conducted with one of Gephi’s creators, Mathieu Jacomy. The interview was meant to fill in the gaps that the creator has sole knowledge of - the history and intended use of the tool, its development, relationship with the field, the creator’s considerations of the current research practices etc. A creator of the tool may be considered also an expert user and a researcher, highly knowledgeable about the method that they are implementing through the instrument, therefore, his views are important for framing the overall discussion.

2.1.1. Interviews with researchers

Interviewees and general format.

The interviews (40-60 min.) were conducted face-to-face with 6 expert Gephi users27 of varying levels of experience, who were chosen due to having conducted work that utilises Gephi for academic purposes. The interviewees varied from MA students to senior researchers in their academic experience within the field of new media studies (and one interviewee from communication science) in order to account for various experiences and opinions. All interviewed researchers resided in the

26_{See: Rieder and Borra}

27_{Prior to starting the interviewing process, 2 pilot interviews were conducted in order to receive feedback, improve questions and}

(31)

Netherlands. Despite being fully aware that this is not a representative sample and the findings cannot be generalised to the overall field, the thesis aimed to explore differing research processes.

As for the interview format, a semi-structured approach was chosen. According to Bernard, it is effective when there is not a chance to interview people more than once and the time limits are important. Similarly to unstructured interviews, which are more common in long-term ethnographic fieldwork, semi-structured interviews are based on a plan, but engage in very little control over the respondents’ answers. A semi-structured interview ‘[d]emonstrates that you are fully in control of what you want from an interview, but leaves both you and your respondent free to follow new leads’ (Bernard 212).

For a part of the interview an elicitation interviewing method (Johnson and Weller 491) was followed. That is, a supportive material (a ‘prop’) was used for a part of the interview, which the respondents prepared for prior to the meeting. The interviewees were asked to either revisit a specific data sample they have used for their own research or were provided with a sample Twitter dataset in a Gephi-supported file format. They were requested to briefly interact with it and take notes on what they did (or, in case of using their own previous research - think back to the process of it). This step was used to start the conversation more easily as well as gain more sincere initial insights into how researchers approach the use of the instrument. Additionally, it was useful for grounding the interview in social media research, focusing the attention on a quite specific dataset and giving the interview a starting point, as opposed to jumping from one case to another. It needs to be noted, however, that the use of the supportive materials cannot in any way be equated with a participant observation for being very controlled and highly artificial. On the other hand, it does provide a certain observational element to the current analysis. Moreover, it makes available more freedom for the interviewee to describe a particular way they conduct research with the tool as opposed to the interviewer prescribing particular, sometimes-biased, opinions of how it should be done.

Interview process and structure.

After starting the interviews from a verbal informed consent statement28, each of them encompassed three bigger parts, which were organised into topics in relation to their

28_{The verbal consent discussed the following topics with the researcher: the purpose of the research, a clarification of it not being a}