Deep networks as associative interfaces to historical research

TY - CHAP

T1 - Deep networks as associative interfaces to historical research

AU - van den Heuvel, C.M.J.M.

AU - van Vugt, Ingeborg

AU - Van Bree, Pim

AU - Kessels, Geert

PY - 2020

Y1 - 2020

N2 - The most widely used software tools for network analysis have the explicit goal of creating patterns that visualize features of the underlying data that are regarded as representative for answering research questions or testing hypotheses. Their underlying algorithms often support quantitative analyses and visualization of large data sets. In general statistical methods are predominantly used to explain to which extent these visualizations are representative for the underlying data. This contribution is based on the assumption that full data integration, in particular in the humanities, is impossible for reasons of incompleteness, complexity, ambiguity and uncertainty in data. Therefore the focus should not be on pattern recognition in combination with statistical methods of network representations alone. We need to include approaches that allow users to explore and to interact with these incomplete and complex data. In short, we do not need just networks as representations but also networks as interactive interfaces. These interfaces must enable users to explore, to interact and to make associations with their own selections of data that can combine data-driven and research question-driven approaches. To this end we first discuss problems experienced within the data-driven project Circulation of Knowledge/ePistolarium of labelling automatically generated topics. We focus on the tension between explicit and implicit terms that biased our interpretations of network representations of our research case dealing with the role of confidentiality in the correspondences present in ePistolarium. This discussion is followed by a brief exploration of potential solutions to overcome the previously mentioned tension in representing computer generated terms in the future. These introductory paragraphs on large scale data-driven research are followed by a discussion of the experiences with a small scale experiment Mapping Notes and Nodes: Exploring potential relationships in biographical data and cultural networks in the creative industry in Amsterdam and Rome in the Early Modern Period, in which a bottom-up approach was followed to create manually multi-layered historical, intellectual and technological networks with the software application Nodegoat. These discussions of experiences with the development of computer-generated and manually created historical networks are followed by an exploration of their implications for user interfaces of historical network research and by an introduction to the term “deep networks.” After a discussion about deep networks in relation to our experiences with the Circulation of Knowledge/ePistolarium and the Mapping Notes and Notes projects we contextualise the term by similar notions in literature in order to explore its potential in the wider contexts of historical network research and digital humanities. We return to our experiences with mappings of unstructured and structured data by introducing in a more generic way classifications of terms that change in meaning over time (concept-drift) and visualization of uncertainty. We claim that current experiments with the representation of big data from multiple perspectives can be read as computer-assisted deep networks that can be used to create associative interfaces to historical research networks similar to the manually created deep networks of the Nodegoat tool. We conclude that the creation of such deep networks not only can contribute in bridging the gap between unstructured and structured data, between distant and close reading, between qualitative hermeneutic approaches and quantitative statistical methods but holds a promise for future historical network research and digital humanities at large.

AB - The most widely used software tools for network analysis have the explicit goal of creating patterns that visualize features of the underlying data that are regarded as representative for answering research questions or testing hypotheses. Their underlying algorithms often support quantitative analyses and visualization of large data sets. In general statistical methods are predominantly used to explain to which extent these visualizations are representative for the underlying data. This contribution is based on the assumption that full data integration, in particular in the humanities, is impossible for reasons of incompleteness, complexity, ambiguity and uncertainty in data. Therefore the focus should not be on pattern recognition in combination with statistical methods of network representations alone. We need to include approaches that allow users to explore and to interact with these incomplete and complex data. In short, we do not need just networks as representations but also networks as interactive interfaces. These interfaces must enable users to explore, to interact and to make associations with their own selections of data that can combine data-driven and research question-driven approaches. To this end we first discuss problems experienced within the data-driven project Circulation of Knowledge/ePistolarium of labelling automatically generated topics. We focus on the tension between explicit and implicit terms that biased our interpretations of network representations of our research case dealing with the role of confidentiality in the correspondences present in ePistolarium. This discussion is followed by a brief exploration of potential solutions to overcome the previously mentioned tension in representing computer generated terms in the future. These introductory paragraphs on large scale data-driven research are followed by a discussion of the experiences with a small scale experiment Mapping Notes and Nodes: Exploring potential relationships in biographical data and cultural networks in the creative industry in Amsterdam and Rome in the Early Modern Period, in which a bottom-up approach was followed to create manually multi-layered historical, intellectual and technological networks with the software application Nodegoat. These discussions of experiences with the development of computer-generated and manually created historical networks are followed by an exploration of their implications for user interfaces of historical network research and by an introduction to the term “deep networks.” After a discussion about deep networks in relation to our experiences with the Circulation of Knowledge/ePistolarium and the Mapping Notes and Notes projects we contextualise the term by similar notions in literature in order to explore its potential in the wider contexts of historical network research and digital humanities. We return to our experiences with mappings of unstructured and structured data by introducing in a more generic way classifications of terms that change in meaning over time (concept-drift) and visualization of uncertainty. We claim that current experiments with the representation of big data from multiple perspectives can be read as computer-assisted deep networks that can be used to create associative interfaces to historical research networks similar to the manually created deep networks of the Nodegoat tool. We conclude that the creation of such deep networks not only can contribute in bridging the gap between unstructured and structured data, between distant and close reading, between qualitative hermeneutic approaches and quantitative statistical methods but holds a promise for future historical network research and digital humanities at large.

KW - Network analysis

KW - historical networks

KW - interfaces

KW - deep learning

KW - deep networks

M3 - Chapter

SN - 978-1-138-73130-1

T3 - Digital Research in the Arts and Humanities

SP - 189

EP - 223

BT - The Power of Networks: Prospects of Historical Network Research

A2 - Kerschbaumer, Florian

A2 - von Keyserlingk-Rehbein, Linda

A2 - Stark, Martin

A2 - During, Marten

PB - Routledge

CY - Abingdon UK New York US

ER -