LinkedArt: exploring network analysis in art history

by Sophia Renz and Vanessa Tissen

The beginning

It all started with the seminar on network analysis in the summer semester of 2020. After learning about the basics of network theory and building networks in Python ourselves, the teachers Aline Deicke and Demival Vasques Filho asked us students to work in groups to develop a project combining our individual humanities backgrounds with network analysis. We are specialists in art history, which we wanted to include in the project. On top of that, the IEG DH Lab provided us with funds and support to further explore the application of network analysis in the field, e.g. whether art history datasets are available and to what extent they are usable or which art historical analyses or topics have already been done. The research project was kept relatively open, so we were able to look at the subject matter first. Tasks and questions developed during the following research.

Literature review/state of the art

Several recent collections of essays and books, such as „The Routledge Companion to Digital Humanities and Art history“ (New York 2020), or „The Network Turn“ (Cambridge 2020) show an increased interest not only in digital methods but also specifically in network analysis in the humanities and furthermore in art history. It shows in further published papers that there are mainly two fields of interest based on the type of data the researchers are analyzing: the status of artists or bidders (with data of the art market, museums, auctions) and artistic collaborations (with data from primary source documents, e.g., letters, manuscripts, records). Maximilian Schich had a different approach using stylistic analysis with databases like ‘Census’ or ‘Zuccaro’ and was mainly looking towards the art objects. Schich worked with scalable image matrices to approach the question of canon in art research. His dissertation „Rezeption und Tradierung als Komplexes Netzwerk“ from 2007 and his following and ongoing work is a big pool of a meanwhile widespread approach towards network analysis for art history and generally the field of culture.

We found two noticeable ways in which the literature deals with network data. First, by applying social network analysis methods, without advancing these methods themselves, as a desire to expand the use of digital methods in art history. Second, by visualizing large networks, which requires a close connection with large database projects, such as ‘Cornelia’, ‘Census’, ‘RED’, or museum data sets. These data are generated from art sales and auction catalogues, or edited correspondence and other primary sources. The effort required to create the datasets is described as very laborious and time-consuming, especially when evaluating and compiling primary sources, because some projects require the inclusion of an extraordinarily large number of sources, or also because data sets must be very cleanly prepared for the visualization tools (e.g. Gephi) that are often used. This small literature review already shows that a high sensitivity for data sets is necessary in order to analyze/calculate really meaningful networks and not to fall too much into purely statistical analyses. This also includes taking into account the critical relationship to data often discussed in the literature under consideration with regard to its completeness, one-sidedness or temporal dimension.

Working process finding questions and changing structure

As newcomers to digital humanities and especially network analysis we initially found the process of an analysis difficult. On the one hand, because we were unsure how to combine art history and network analysis or whether network analysis can be helpful for art history at all. On the other hand, for the lack of experience on how a network analysis actually works, which data sets are suitable for it, which basic requirements must be met and how the output is interpreted. These and many more thoughts/questions accompanied us throughout the whole project.

Because of these insecurities we choose the liberal arts way of doing things: reading reports to approach network analysis. We could see what the current state of research is, what data sets researchers have worked with, what questions are possible and what results can be read from the output. But after trying to find meaningful data sets and questions, we recognized that this procedure is useful for liberal arts but not absolute for the digital humanities: sometimes it makes sense to leave the theoretical part behind, just try it out and not worry about failure. Sometimes the practical approach is more helpful in understanding the basic principles of processes and theories.

Then, we changed the way of thinking and working. We detached ourselves from the idea of discovering something completely new and tried out simple commands and methods with the help of our supervisors.

Results and thoughts about the MoMA data set

After a few detours, we started dealing with the data sets of MoMA (Museum of Modern Art) about their exhibitions and artworks. Since one of the datasets contains all MoMA exhibitions from 1929 to 1989 with artist and curator information, it was possible to investigate how often and when which artist was exhibited with whom. For this purpose, a two-mode network between the ArtistIDs, CuratorIDs and the ExhibitonIDs was used, which was extended by attributes such as their labels, dates or nationalities. Due to network projections is it possible e.g. to analyze the behavior and connections between the artists. The whole analysis happened in Python with the packages networkx and pandas.

Fig. 1. Degree distributions of both sets of nodes of the two-mode network (MoMA data set) connecting artists to exhibitions. The degree of an artist represents the number of times exhibitions in which the artist was exhibited. The degree of an exhibition represents the number of artists being exhibited. Most artists were exhibited just a few times, but the are some exhibited over a hundred times. For the exhibitions, the distributions is a little more balanced, with the number of exhibitions decreasing slower with the number of artists appearing in them.

 

Admittedly we are still at the beginning of the analysis and we need to dig deeper into the data set. But by contrasting which artists were exhibited the most in each decade, we identified some trends and patterns in terms of curators‘ exhibition decisions. Conversely, if the concepts of the exhibitions are understood, “advice” can be formulated for artists on how to exhibit more frequently. (It should be noted hat unfortunately the data sets are only available up to 1989, so no current assumptions can be made!)

You will be more exhibited, if …

  1. … you cover different styles in your work. For example, if you do Cubism, Surrealism and Futurism you increase the chances of being exhibited at shows of these art styles. It is also crucial to maintain contacts with fellow artists: Sometimes it is easier to become exhibited as a member of a group.
  2. … you work with different techniques. As with art styles, you can increase the possibility of being exhibited if you photograph, paint, print, design architecture or clothing, illustrate, and make sculptures. This way you can offer artworks at various exhibitions. It could also be helpful to be productive and active as an artist for a long time.
  3. … you are living and working in different countries. European artists like Marc Chagall, André Breton, Max Ernst, Fernand Léger and Max Beckmann, which immigrated to the USA before or during the Second World War had a great direct influence on European as well as American art, especially to the American Abstract Expressionism. At the end, these are considered American artists and are part of the exhibitions about both European and American artists.
  4. … you take reference to other famous artists as heirs or following artists. When the museum is planning an exhibition on Paul Cézanne, they could also talk about his influence on Pablo Picasso, Georges Braque, or Henri Matisse, and exhibit them as part of that. This also works in the other direction: You will be more exhibited, if other artists are taking references to you, when you are like a pioneer for other artists and styles. While Jasper Johns and Robert Rauschenberg cannot be attributed to Pop Art, they have had a significant impact on this style of art and are shown at most Pop Art exhibitions.
  5. … you fit into the education concept. Meet the style criteria to become a main representative of the style. As a result, you will not only be referred to more often as an example of a style, but there is a chance that you will rise to become a crowd puller.
  6. … you have influential sponsors who collect your art. This increases the chance that their collection will be exhibited in major museums or that the collectors will donate your works to a major museum. Since most American museums are private, it is more profitable for them to display their own artwork than to loan artwork.
Fig. 2. Network of co-exhibited artists in the MoMA data set. Node colors represent the nationalities of the artist according to the legend. Node size is proportional to the degree (number of co-exhibited artists) of the artist. Artists that are named have the highest betweenness centrality, meaning that they are potential bridges between different styles. Although the vast majority of artists are American, there are central artists from several other countries.

Conclusion

By turning to the MoMA exhibition data set, it became clear that the connection of artists and exhibitions/curators brings an interesting constellation for a two-mode network analysis in which we were able to gain a lot of experience. It was important to deepen network analysis  with the coding language Python (notably with the networkx package). By not using visualization tools, the data processing and analyses of multi modal networks are more dynamically adaptable and more versatile analyses and visualizations can be implemented. It is vital to be always aware of the complexity of a two-mode network: not every calculation that can be carried out by the computer supports a meaningful conclusion.Furthermore, the inclusion of general statistics can give profitable additional information, which, however, should not become the center of the analysis.

Our approach of searching for previously unseen constellations/effects/knowledge about the underlying data exposed the need to compare the results with art-historical knowledge, knowledge that is at best already available at the time. In our analysis, many questions remained unanswered by missing out the foundation of deepening research on the specific topic of exhibitions and their artists. As well for the network analysis itself further calculations would have been possible particularly with a theoretical foundation, but not enough time was left to follow up broader ideas. To be more convincing in our analysis, we would have to investigate further exhibition data from other museums, and even then, many special factors (e.g. economics) of the ‘private-museum-world’ that would play a major role in the evaluation of the analysis results must be considered. Generally speaking, this means even more that a collaborative approach of specialized researchers from different fields is a good way to work on these new research questions and analyzing methods.

Finally, it should be said that network analysis as a method of research in art history should by no means be written off. On the contrary, in our eyes quantitative analyses and especially network analyses can be profitable in art history. If art historians are sensitized to consider their conglomerates of primary sources more often as complex data sets for network analysis, collaborative network analysts could still amplify research questions and topics.


Cite this article as: Sophia Renz and Vanessa Tissen, "LinkedArt: exploring network analysis in art history," in Digital Humanities Lab, 24/02/2021, https://dhlab.hypotheses.org/?p=1867.

About the authors

Sophia Renz

  • Ausbildung
    • Johannes Gutenberg-Universität Mainz (JGU)/Hochschule Mainz (HS)
    • Université de Paris-Sorbonne (Paris IV), Erasmus-Auslandssemester in Paris (2017)
    • Bachelor of Arts in Kunstgeschichte und Musikwissenschaft (JGU, 2014-2018)
    • seit 2017 im Studiengang Kunstgeschichte, Master of Arts (JGU/HS Mainz)
    • seit 2019 im Studiengang Digitale Methodik, Master of Arts (JGU/HS Mainz)
  • Berufserfahrung:
    • Werkstudentin im Hessisches Hauptstaatsarchiv Wiesbaden (2015)
    • Studentische Hilfskraft im Stadtarchiv Ingelheim (2018)
    • Praktikum an der Photothek, Kunsthistorisches Institut in Florenz, MPI (2019/20)
    • Wissenschaftliche Hilfskraft am IKM, Abteilung Kunstgeschichte der JGU Mainz für digitale Lehre (2020)
    • Mitarbeit am Forschungsprojekt „LinkedArt“ am IEG, Mainz (2020)
    • Wissenschaftliche Hilfskraft am IKM, Abteilung Kunstgeschichte der JGU Mainz für Prof. Dr. Elisabeth Oy-Marra und Homepage der Abteilung (seit 2020)

Vanessa Tissen

  • Ausbildung
    • Johannes Gutenberg-Universität Mainz (JGU) /Hochschule Mainz
    • Bachelor of Arts in Kunstgeschichte und Germanistik (2013-2018)
    • seit 2018 im Studiengang Kunstgeschichte, Master of Arts
    • seit 2019 im Studiengang Digitale Methodik, Master of Arts
  • Berufserfahrung:
    • Mitarbeit beim 33. Deutschen Kunsthistorikertag, Mainz (2015)
    • Praktische Tätigkeit in der Galerie Dr. van der Koelen, Mainz (2015-2018)
    • Praktische Tätigkeit in der Galerie Dömland, Mainz (2015-2017)
    • Praktikum in der Kulturabteilung, Mainz (2016)
    • Praktische Tätigkeit in den Opelvillen Museen, Rüsselsheim (2017)
    • Mentorin für Kunstgeschichtsstudierende an der JGU, Mainz (2018-2020)
    • Praktische Tätigkeit im Kunst- und Auktionshaus, Wiesbaden (2019)
    • Praktikum am Kunsthistorischen Institut, Florenz (2020)
    • Mitarbeit am Forschungsprojekt „LinkedArt“ am IEG, Mainz (2020)

Schreibe einen Kommentar

Deine E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert.

Diese Website verwendet Akismet, um Spam zu reduzieren. Erfahre mehr darüber, wie deine Kommentardaten verarbeitet werden.