Contents

Part I: “Do you have to know how to code?”

Part II: Affordances and Limitations of Network Analysis

Part III: Using Gephi to examine Artist Networks

Tools and Methodology

Network Analysis

Conclusion

Bibliography (Part I, II, III)

Tools and Databases used

Part I: “Do you have to know how to code?”

I think to be able to discuss this question, it is important to first analyze the question itself. Starting with “have to”, one can ask who is obliging here and why. The field of Digital Humanities is very broad and encompasses a huge variety of fields, research areas, and professions. If I were to ask myself this question, I would ask if I have to know how to code for my areas of interest.

Secondly, “know how to” is very vague when it comes to programming. There are many different levels of skills – from basic Python to highly advanced Java – and it should be evaluated which level of programming is necessary for which task.

Finally, the biggest question here for me is the meaning of “code”. How many languages does it involve? Do I know how to code if I know SQL, or have basic proficiency in Python, while most tools I need to use require Java? Do I need to be a certified programmer to be able to say that “I can code” or is a course on udemy enough? As Ramsay (2013) rightly wrote:

All the technai of Digital Humanities — data mining, xml encoding, text analysis, gis, Web design, visualization, programming, tool design, database design, etc — involve building; only a few of them require programming, per se.

So far, what I experienced in DH is a great confusion around the topic of coding. Starting with as vague a question as the one above, it is hard to find an answer. I think that everyone involved in DH needs to ask themselves this question about their specific background and areas of interest, and it is not something that can be postulated for the whole community.

On a personal level, I have been asking myself this question many times and I have not been able to find an answer yet. I have some basic knowledge of mark-up languages and I can use PythonTurtle, but that is about as far as my coding skills go. Sometimes I wish I was able to code because I think that it would make my life a lot easier, enabling me to do things myself instead of having to ask someone or use an existing tool. Nevertheless, I have never been stuck because of the inability to use a programming language, and so far, I always consciously made the decision against coding and for learning other skills that help me more with getting where I want to be.

I think that as programming itself gets more and more user-friendly – I heard from my programmer friends that the field might be moving towards intuitive languages like Kotlin and away from complicated, unintuitive languages like Java – it will get easier to learn the skills. I assume that programming will move towards drag and drop and intuitive, user friendly design while pre-programmed tools are already multiplying, enabling people who do not know how to code to create what they wish to.

Looking at DH as an academic field, I completely agree with Ramsay that its distinctive feature is that it involves creation and building, as opposed to mere analysis and extraction like the traditional humanities. This “building” can have multiple forms, and while it surely encompasses programming an App or writing a software, these are not essential skills of a DH scholar. Rather, bridging the gap between digital literacy and traditional humanities knowledge and research is what I see as the responsibility of DH, and I think this is what DH scholars and graduates are valued for. While expectations, especially of young people, regarding tech savvy are high, it seems to be a dangerous fallacy that growing up with technology automatically results in technical proficiency. As O’Sullivan et al. (2015, pp. i143–i144) discovered, “[t]here is a marked distinction between ‘using’ technology and ‘understanding’ technology.”

Learning a programming language is not something one can do with a few hours a week within a year. It takes time and personal investment to excel at it, which should be made clear to anyone attempting this challenge. If programming becomes a requirement for DH scholars, then it has to be taught extensively to students, which will in turn take time and resources from the teaching of traditional humanities skills. Learning how to code and focusing on technical skills can be of no harm for a DH scholar, but I would be careful to postulate it as mandatory and thus take away the focus from traditional humanities skills and knowledge building. As O’Sullivan et al. write (2015, p. i147) “You do not ‘have’ to code, as long as you can work—effectively—with someone who does.”

Personally, my inability to code and my lack of time to learn it is often a cause of anxiety for me, and I know that I am not the only DH student who feels like that. I would say that just because I do not know how to code does not mean I cannot create things and be a good scholar in the field of Digital Humanities. While it is important to have an open mind towards the latest technology and the current requirements of the job market, it is also important to focus on one’s personal skills. This is what I love about DH: the freedom it gives to be creative and use one’s knowledge and skills in manifold ways, and the need for critical thinking and analysis. Contrary to the hard sciences, which are closely defined and prescribed, the beauty of the humanities lies in their interdisciplinary, open-minded, and creative nature. The Digital Humanities are a perfect extension of this, with an emphasis on digital tools, but without clear-cut requirements and prescriptive have-tos. Concluding, if I were forced to answer Ramsay’s question in one sentence, I would probably do so by saying “It depends…”.

Part II: Affordances and Limitations of Network Analysis

With the recent rise in popularity of visualizations and infographics, network visualizations and network analysis have become very popular in the digital humanities and other fields for a variety of reasons. Having an ever-growing range of user-friendly free tools like Palladio, Onodo and Gephi at our fingertips, the challenge is to evaluate which method and which tool is appropriate for the kinds of questions we are asking. Furthermore, in order to avoid misrepresentations and trivial results, it is important to be aware of the affordances and limitations of the computational methods used and what it can and cannot do to answer the research question.

Network analysis has its origins in graph theory and was, in its beginnings, primarily used for mathematical, economical and logistical studies (Frank and Frisch, 1970). In the context of the social sciences, Boissevain (1979, p. 392) calls network analysis:

An analytical instrument which views circles of relatives and friends, coalitions, groups and business houses, industrial complexes, and even nation-states as scatterings of points connected by lines that form networks […]. Network analysis asks questions about who is linked to whom, the content of the linkages, the pattern they form, the relation between the pattern and behaviour, and the relation between the pattern and other societal factors.

In the context of digital humanities, network analysis can be used to examine relationships between literary works, characters, institutions, historical persons, events, and many more. Due to its very easy layout – nodes and their links (edges) – seemingly almost anything can be represented as a network and analyzed accordingly.

For my dissertation, I am constructing a network visualization of artist networks of the 19th century with Gephi in order to analyze the usefulness of this method for research purposes in the field of Art History. I will thus use my network as an example to examine affordances and limitations of network analysis for this specific purpose.

In his reappraisal of network analysis, Boissevain (1979) mentions multiple advantages of this method, most of which can be applied to the digital humanities. Firstly, he states that network analysis focuses attention on interlinkages and highlights interdependency between units of analysis. In my case, this enables me to see links between artists, groups, institutions and movements. Additionally, Gephi allows me to adjust the size and color of the nodes according to different metrics, which makes it possible to detect relative importance as well as the type of node immediately.

Figure 1: Artist Network Visualization – colored by attribute “instance of” and sized by total degree.

Furthermore, network analysis focuses on the content of relations and encourages their investigation. I can see the type of relationship that exists between different units and investigate their relevancy and meaning for the network as a whole.

Figure 2: Types of Edges between Nodes.

Secondly, Boissevain (1979, p. 393) writes that network analysis “brings into sharp sociological focus the difficult analytical category of friends-of-friends, those persons who lie just beyond the researcher’s horizon because they are not in direct contact with his informants.”

In my context, this means that through the visual display of relationships between “actors” in my network, I am able to detect relationships that might have stayed unnoticed because I can see different degrees of relationship – i.e. “friends of friends”. In the digital humanities, networks are often constructed on the basis of literary texts or historical sources, which contain massive amounts of data. Network analysis can help to get an overview of the data and see patterns, degrees of relations, and “hubs” of nodes that are linked numerous times (Sweeny, 2013), while having the potential “to reveal the significance of actors marginalized by canonical narratives of art history through the mining of archival data and also to track unforeseen transnational and intercommunal histories of artistic exchange” (Kienle, 2017, p. 5).

Once the network is created, different software allows for different levels of examination and analysis of details. For example, Gephi makes it possible to enable a timeline, which adds a temporal dimension to the network.

Figure 3: Gephi Interface with Timeline at the Bottom.

Different filters allow to focus on a selection of nodes and their edges and investigate further, which enables different levels of close and distant reading of the network. Jeffrey Drouin (2014, p. 122) writes about an “ego network”, the “immediate network surrounding a single node in the data set, seen in isolation”, which can be helpful when examining the relationships that one particularly influential artist had.

Figure 4: “Ego Network” of James Abbott McNeill Whistler.

Nevertheless, with all these advantages and affordances in mind, it is crucial to understand that a network visualization and its analysis can only be as good as the data that it feeds from. Here lies the most crucial limit of macro-analysis in general and network analysis specifically: any computational method needs data and this data has to be found, compiled and cleaned before it can be of any use. During this process, it is inevitable that some of the information and meaning of the primary material gets lost or distorted. In the example of network analysis, the program can only take into account what can be considered nodes, their attributes, and their relations – basic metadata about units of analysis. Any other information such as personal experiences of the artists or their individual interests that cannot be expressed through relational data is lost. For me, the biggest limitation is that network analysis can only show connections that I discover and decide to add to the data set.

Network analysis cannot describe processes and reasons for relations between units, which can – depending on the purpose of the analysis – be crucial for understanding and might even lead to misinterpretation. My network cannot show the reason for the relationships between the artists and it also does not show the relevancy of these relationships. Just because an artist was another artist’s student does not mean that s/he was influenced by the teacher – this can only be an assumption. Furthermore, as Boissevain  (1979, p. 393) warns, this method all too often produces trivial results when it is used for the wrong motifs:

The concern with method, classification, and networks-as-things-in-themselves, rather than with the ideas and problems that the practitioners are attempting to solve, characterizes not only the results but also, alas, the way in which those results are reviewed.

Thus, it is important to always keep the research question in mind while using network analysis and to stay away from easy and obvious conclusions.

Finally, Miriam Kienle (2017, pp. 5–6) emphasizes that the biggest danger of network analysis is that, while it has great potential to be beneficial, by revealing undetected connections and marginalized actors:

It may also paradoxically silence social hierarchies and mechanisms of marginalization, as well as historical disruptions to them, if the principles underlying the data are not interrogated from the outset.

This is mainly due to the cultural, political and economic dominance of Western Europe in many disciplines of the humanities.

Concluding, although there are limits and dangers of network analysis as a computational method for the digital humanities, in the right context and when used to answer the right questions, this tool can be of great benefit and while it might not always provide the desired answers, it definitely produces insights and potential further questions by making visible what would otherwise stay undiscovered.

Part III: Using Gephi to examine Artist Networks

As mentioned above, I already created a provisional data set of 19th century French Impressionist artists and their connections for my dissertation and executed a visualization with Gephi (the dataset can be found at https://github.com/ckdigitalarts/network). In the following I will use this visualization and Gephi’s network analysis tools as well as the interactive browser display with the help of sigma.js to address the following questions, relevant to research in the field of Art History, which will be used in my dissertation to demonstrate the value and benefits of network visualization and analysis for this specific academic field: Is there a correlation between total degree of connections and popularity of artists? Can statements be made about the role that centrality plays for the Impressionist artist movement as opposed to the academic tradition?

Tools and Methodology

The dataset comprises 120 nodes and their attributes and 203 edges between those nodes. All data was collected from an existing data source, Wikidata, and processed as two CSV files, which were then imported into Gephi to produce a directed graph. Nodes are divided into five instances: human, movement, group, organization and institution (Figure 5). Edges have different labels like “student of”, “founded by”, and “educated at” that specify the type of connection between individual nodes. Using the SigmaExport plugin and the sigma.js Javascript library for Gephi enables me to display the network as an interactive visualization in my browser which makes it possible to select individual nodes and view their attributes at the same time (Figure 6).

Figure 5: Network visualization in Gephi colored by attribute “Instance of” with edge and node labels.

 

Figure 6: Browser view of network with sigma.js (See http://network.ckdigitalarts.com/gephi/).

In order to determine the popularity of artists, the number of search results for a simple search in the online art history Database Oxford Grove Art Online (http://www.oxfordartonline.com/groveart) will suffice for this essay to compare popularity and total degree of connections.

Network Analysis

Gephi allows to size nodes by different metrics, one of them being total degree – the total amount of in and out connections. Displaying the network this way with a ranking of nodes by size by degree (min size 2, max size 20, linear distribution) shows three “hubs” – nodes with a great degree of edges going in and out (Sweeny, 2013, p. 218). These hubs are the artists Jean-Leon Gérôme, his teacher Marc-Charles Gabriel Gleyre, and the famous French art school École des Beaux-Arts (Figure 7). As I am focusing on artists, and thus nodes that have the attribute “instance of human”, the École des Beaux-Arts is not relevant here. Additionally, for matters of simplification and clarity, the acronyms JLG and MCGG will be used from here on for above artists.

To determine the popularity a search in the art history database Oxford Grove Art Online was performed with the following results:

Artist

No. of results

JLG

104

MCGG

7

Oxford Grove Art Online search results for “Jean-Léon Gérôme”, and “Marc-Charles-Gabriel Gleyre) [search performed 16/03/2019].

While the size of the node JLG correlates with a relatively high score of search results, the size of MCGG is surprising when looking at the search results and suggests that the degree of a node does not allow assumptions about the artist’s popularity in art historical writings (Figure 7). This claim is strengthened by the fact that well known – and much written on – impressionist artists like Edgar Degas and Paul Cézanne, who return a high number of search results, do not show a high degree of edges in the visualization (Figure 8, Figure 9).

Artist

No. of results

Claude Monet

126

Paul Cézanne

252

Oxford Grove Art Online search results for “Claude Monet” and “Paul Cézanne” [search performed 16/03/2019].

 

Literature on MCGG explains that the Swiss artist was largely underrepresented in art historical writing for a long time due to his refusal to associate with the French fine arts academy and his work as “an independent not openly involved with any artistic camp or style” (Hauptman, 1981, p. 17). Thus, it is possible that the visualization allows for a representation of artists and their influence that is less influenced by politics and ideologies.

Figure 7: Graph in Gephi sized by total degree

 

Figure 8: Edges of Claude Monet in browser view

 

Figure 9: Edges of Paul Cézanne in browser view.

 

 

 

 

 

 

 

 

Until the 19th century, the art world was strictly regulated by authorities, traditions, and confined spaces, which controlled who, when, and where art was to be created, published, and distributed. With the decline of the academies and salons in France and the liberation of the art market, artists had to form groups and networks in order to exhibit their work and find supporters. In their study of Impressionist art, White and White (Finkelstein, 1966) identify the Impressionists as the instigators of this change, who opened up the art market for true creativity and innovation as opposed to the academies’ holding on to mass-production of traditional art.

The representation of the network according to betweenness centrality allows to examine the degree to which the traditional hierarchical structure of the academy (few highly influential actors) played a role in Impressionism. Sizing and coloring the network according to betweenness centrality, with darker green and bigger size showing higher scores, makes it possible to examine the structure of Impressionist networks. The visualization displays the traditional academic artist Jean-Léon Gérôme with a great degree of centrality, surrounded by nodes with low centrality, while Impressionism, the artist Claude Monet, and the Société anonyme des artistes peintres, sculpteurs, et graveurs, surrounded by the artists Francesco Filippini, Ludovic-Napoléon Lepic, Edgar Degas and Frédéric Bazille are interconnected among each other with raised levels of betweenness centrality. This can be taken as an indicator for the movement from rigid hierarchies of the early 19th century towards the establishment of small, influential, interconnected groups, driven by innovative artists like Monet and Lepic, that built interwoven, dynamic networks.

Figure 10: Graph in Gephi colored and sized by betweenness centrality.

Conclusion

Concluding, this short study exemplifies that network visualization and analysis can show aspects of art history that might otherwise be underrepresented. An analytical examination and case study of the correlation between popularity in the literature and connectedness in the visualization offers insights into under-researched areas and can open up new research questions. Furthermore, the macro-analysis of artist networks and their centrality indicates a change in structure of social networks among artists with the rise of Impressionism. This remains to be studied further in order to make general assumptions about the nature of the change and its reasons and consequences. Finally, what has to be kept in mind at all times is that if “data visualizations are simplifications, when they become the primary way to communicate research, the rich complexity of research is reduced and only a partial story is told” (Kennedy and Hill, 2016, pp. 776–777). Network analysis, as other computer-assisted technology, can always only be used as supplementary tools for research and require in-depth qualitative analysis of the research subject.

Bibliography (Part I, II, III)

Boissevain, J. (1979) ‘Network Analysis: A Reappraisal’, Current Anthropology, 20(2), pp. 392–394.

Drouin, J. (2014) ‘Close- and Distant-Reading Modernism: Network Analysis, Text Mining, and Teaching the Little Review’, The Journal of Modern Periodical Studies. (Special Issue Visualizing Periodical Networks), 5(1), pp. 110–135. doi: 10.5325/jmodeperistud.5.1.0110 [accessed 2019-03-07].

Finkelstein, S. (1966) ‘Reviewed Work: Canvases and Careers: Institutional Changes in the French Painting World by Harrison C. White, Cynthia A. White’, Science & Society, 30(2), pp. 238–241.

Frank, H. and Frisch, I. T. (1970) ‘Network Analysis’, Scientific American, 223(1), pp. 94–105.

Hauptman, W. (1981) ‘Gleyre, Vernet, and the Revenge of “Les Brigands Romains”’, The Bulletin of the Cleveland Museum of Art, 68(1), pp. 17–34.

Kennedy, H. and Hill, R. L. (2016) ‘The Pleasure and Pain of Visualizing Data in Times of Data Power’, Television & New Media, 18(8), pp. 769–782. doi: 10.1177/1527476416667823.

Kienle, M. (2017) ‘Between Nodes and Edges: Possibilities and Limits of Network Analysis in Art History’, Artl@s Bulletin, 6(3: Visualizing Networks: Approaches to Network Analysis in Art History), pp. 4–22.

O’Sullivan, J., Jakacki, D. and Galvin, M. (2015) ‘Programming in the Digital Humanities’, Digital Scholarship in the Humanities, 30(1), pp. i142–i147.

Ramsay, S. (2013) ‘On Building’, web.archive.org [accessed 2019-02-03].

Sweeny, R. W. (2013) ‘Complex Digital Visual Systems’, Studies in Art Education, 54(3), pp. 216–231.

Tools and Databases used:

Gephi: https://gephi.org/

Sigma.js: http://sigmajs.org/

Github: https://github.com/

Oxford Grove Art Online: http://www.oxfordartonline.com/groveart

Wikidata: https://www.wikidata.org/