Watch the seminar here

AI generated summary of the talk:

This was the fifth and final talk in the HNR online seminar series, where Malte Vogl presented his work on networks of collaboration and interdisciplinarity in scientific institutions. The talk grew out of a large research project on the history of the Max Planck Society from 1948 to 2002, which combined archival digitization, research data infrastructure, and historical analysis. Malte discussed how digital and network-based methods can support the history of science, but also emphasized the methodological difficulties involved in turning large bibliographic and archival datasets into historically meaningful evidence.

The central concept of the talk was the idea of socio-epistemic networks: networks that connect social actors, material representations of knowledge, semantic structures, and environmental conditions. Malte argued that knowledge should not be treated as an abstract object, but as something carried by people, embedded in institutions and materials, expressed through language, and shaped by wider infrastructures. He distinguished between micro-level historical research, macro-level analysis of large datasets, and the more difficult meso level, where one tries to explain large-scale patterns through historically grounded assumptions about individual actors, institutions, and practices.

A major part of the presentation focused on the challenges of “big data” history. Malte showed that bibliographic databases such as Dimensions or OpenAlex are never neutral sources. They contain geographical, linguistic, disciplinary, and temporal biases, and their changing structure creates problems for reproducibility. To address this, he described efforts to create reproducible snapshots of OpenAlex data and to reduce large datasets to the parts needed for a specific historical question.

Malte then introduced two methodological approaches developed from the Max Planck Society project. The first examined collaboration patterns through ego-network visualizations of individual institutes and their institutional partners over time. These visualizations made it possible to compare different institutes and sections of the Max Planck Society, showing, for example, how collaboration patterns differed between fields dependent on large instruments and those based on more easily shared laboratory technologies. The second approach focused on interdisciplinarity. Using OpenAlex subfield classifications, Malte constructed co-subfield networks and compared the Max Planck Society with global publication patterns. This produced “delta networks” that highlighted combinations of fields unusually prominent, or unusually absent, within the Max Planck Society. He emphasized that such patterns are not explanations in themselves; they are starting points for dialogue with historical experts and archival evidence.

In the final part of the talk, Malte presented ongoing work on the history of nuclear fusion research and science diplomacy, in collaboration with Roberto Lalli. This project examines how fusion research contributed to European scientific cooperation by combining social networks of commissions and actors, publication data, and semantic analysis. The talk concluded by stressing that network methods can reveal meaningful macro-level patterns, but that these patterns must be interpreted through expert knowledge, attention to bias, and careful movement between quantitative and qualitative historical analysis.

The Q&A section focused mainly on organizational network history, methodological robustness, semantic networks, and the role of networks in Malte’s historical thinking.

1. Organizational network history and the scale of the method

One participant commented that historical network research has not often focused on the history of organizations, though he recalled earlier work mapping projects funded by the Swiss National Science Foundation. Malte agreed that organizational network history remains less common and explained that his approach is difficult to apply in a small individual project. It requires substantial resources, technical infrastructure, expert knowledge, and time. He also noted the practical importance of data access: proprietary databases such as Dimensions can be useful, but limited access makes reproducibility difficult, which is why he is increasingly interested in OpenAlex despite its imperfections.

2. Sensitivity, robustness, and the role of individual actors

Another participant asked how Malte accounts for the possibility that large shifts in the “delta networks” might be caused by a few individuals joining or leaving an institution. Malte answered that this is an important issue and that sensitivity testing is a necessary next step. One possible strategy would be to compare the observed networks with random networks that preserve similar edge distributions, in order to test whether apparent changes are robust or spurious. He also noted that the statistical assumptions behind the current method are imperfect, because subfields cannot really be treated as fully independent. For Malte, this question also pointed to the need to move between macro, meso, and micro levels: large-scale network patterns may identify historical questions, but specific explanations may require returning to individual institutes, directors, publications, or actors.

3. Semantic networks and the problem of “atoms” of knowledge

Another participant asked about the semantic dimension of the project: whether it is possible to analyze the actual content of scientific papers and trace conceptual innovation through networks. Malte answered that this is one of the hardest parts of the work. The basic problem is deciding what the “atom” of knowledge should be. It could be a word, a group of words, a topic, a semantic embedding, or a more complex mental model. He explained that he had experimented with different ways of atomizing semantic content but did not find them fully convincing. As a result, he has moved somewhat away from treating semantics as a straightforward network layer. Instead, he sees semantic analysis as a difficult field where computational methods may need to be combined with close reading, historical interpretation, and careful attention to how language changes over time.

4. Networks as concept, visualization, method, and motivation

The final discussion returned to a broader question: what role do networks play in Malte’s work? Zef Segal noted that network analysis often moves between several meanings of “network”: a conceptual framework, a visualization, an interpretive tool, and a quantitative method. Malte answered that the aspects that currently interest him most are, in some ways, the least straightforwardly network-like: semantics and the meso level of explanation. He was especially interested in agent-based modeling because it can bring together heterogeneous social actors, different historical roles, emergent activities, and the transmission of knowledge. For him, the value of network thinking lies not only in producing graphs, but in forcing historians to confront the assumptions, biases, and hidden structures embedded in their data, algorithms, and interpretations.