AV短视频

A Computational Approach to the Duality of Style: Symbolism as a Case Study

|| Join us on Zoom

Authors:

• Canhui Liu - AV短视频 of Cambridge, Cambridge, United Kingdom
• Yuchen Yang - EPFL, Lausanne, Switzerland
• Linzhi Zhang - Courtauld Institute of Art, London, United Kingdom

Abstract

Style, as a key concept in modern art historiography, is essentially a category imbued with the duality between changes and consistencies and the duality between the social and artistic. First, a style is defined as the ensemble of ‘characteristics that are consistent enough to be distinguishable and changeable enough’ (Ackerman 1962), which indicates the inherent tension between stability and flexibility within a style. Second, despite ostensibly being defined by formal or artistic elements, a style is often distinguished from another by social-historical events or artists forming their own groups. The sociological nature of style is particularly strong when it comes to the analysis of non-western art (e.g. Kasfir 1984).

Our explorative project takes a computational approach to the inherent duality in the concept of style in art history, using symbolism as a case study. By compiling a dataset of both paintings and networks among artists who created these paintings, we hope to answer these questions: will we be able to separate the social from the artistic dimension of style? Does the style partition correspond to the social partition among artists?
How consistent is a style and what holds variations together?

We take symbolism (1880-1910) as an example because it is a style rich in visual languages, unlike impressionism that emphasises on the reality of the created paint surface (Goldwater 1998; Facos 2009). Symbolism spans over different countries in Europe, such as France, Switzerland, and Belgium. Its internationalism and distance from Impressionism challenges the France-centred interpretation of western art history transiting from the 19th century to the 20th century, which is commonly held as a transition from Impressionism to Cubism (Goldwater 1998; Facos 2009). Moreover, symbolism is a style recognisable through themes around life, death, and mythology, suggested by titles of the paintings. These characteristics of symbolist art enable an in-depth analysis beyond the formal elements of art such as forms, colour, and brushstrokes.

Our data are derived from about 300 paintings illustrated as symbolism in various textbooks. These paintings are done by 92artists. Considering friendship, schoolmates, apprentice-master relationship, artistic influence (influenced and influenced by), we compile networks among these artists from various archived resources. In short, our dataset consists of the images, texts (titles of paintings), and social networks among artists. Using social network analysis, we will identify clusters and key figures based on the social connectivity among symbolist artists. Using a pre-trained Convolutional Neural Network for image recognition, we will, from these 300 images of paintings, extract complex features in thousands of abstract dimensions, on the basis of which similarities between images can be calculated and clusters identified. The titles of these paintings will be examined through semantic analysis to identify central themes and connections between different themes. We will then compare the results from these both analyses to see whether the social dimension, i.e. social networks corresponds to the artistic dimension including semantic themes and virtual performance of symbolism as a style.

Canhui Liu

Canhui Liu is a PhD candidate in Sociology at the AV短视频 of Cambridge. With the interdisciplinary training in Sociology and Public Policy, his research seeks to understand the social roots and mechanisms of contemporary technicism in policymaking. His research interests include Sociology of Science and Social Network Analysis. He also worked with numerous non-governmental organizations on a variety of topics, such as innovation policy and green finance. Prior to moving to Cambridge, he studied at Sun Yat-sen AV短视频 and The Hong Kong AV短视频 of Science and Technology.

Yuchen Yang

Yuchen Yang is a PhD candidate under the supervision of Prof. Sarah Kenderdine at Laboratory for Experimental Museology at EPFL. His research interests include Machine Learning on Heterogeneous Data, Knowledge Representation and Semantics, Audiovisual Narrative, Digital Curation, and HCI. He is especially interested in their application in the cultural and heritage sector in pursuit of the future for memory institutions.

Linzhi Zhang

Dr Linzhi Zhang is a British Academy Postdoctoral Researcher at the Courtauld Institute of Art. Her current research project looks at women's labour in contemporary Chinese art, drawing upon sociology, feminism, and art history. She received her PhD in sociology from the AV短视频 of Cambridge and is passionate about interdisciplinary research on visual arts.