A New Way of Looking: Tipping Points in Categorization and Canonization in the Art Market (2020-2025)

A New Way of Looking: Tipping Points in Categorization and Canonization in the Art Market (2020-2025). Thanos Efthymiou (PhD candidate) – supervisory team: Monika Kackovic (FEB), Stevan Rudinac (FEB), Nachoem Wijnberg (FEB), Marcel Worring (FNWI).

(2019) Digital Transformations ABS/FEB grant of € 200,000 [Granted].

In this project, we analyze a large corpus of fine art paintings - spanning the Early Renaissance to

Contemporary Art - to study artistic and economic success in the art market. Using advanced computer

vision techniques, we measure - at the level of the pixels per painting - the visual information encoded

in 60,011 publicly available fine art paintings created by 942 renowned artists. Furthermore, we develop machine learning techniques to analyze both the visual and the textual data structure to predict artists’ success along multiple performance indicators of recognition, popularity and historical importance, such as Wikipedia mentions, Google Ngram, Google Trends, peer influence networks, art reviews, exhibition history, and art auction sales. In doing so, we gain a deeper understanding of the differential dynamic processes that unfolded over time and give provenance to Western cultural heritage as we know it today, as well as advancing a fine-grained understanding of the (current) economic drivers of the art market.

Key Points

• We jointly model visual and semantic information;
• The scope of our data analysis is unprecedented;
• Our state-of-the-art algorithms significantly outperform other methods;
• We focus on success in the art market as being multifactorial, transcending mere features of

artworks;

• Our multi-modal methods reveal trends and relations that are difficult to discern using singular

approaches.

Proprietary knowledge

Our newly developed machine learning algorithms, which are agnostic to any fine art information,

allow us to understand and predict artistic success by:

• analyzing a large corpus of visual images (at the pixel level) to identify and measure artists’
• relative level of originality and its influence on different dimensions of success;
• identifying relationships between artists and paintings vis-à-vis hard-to-discern influence

relation;

• understanding the development of artists over time by analyzing the temporal sequences of
• paintings in artists’ careers, which helped estimate success (especially for new artists).

Business Applications:

The implications are manifold, particularly for business. Note: the below possibilities are not turn-key

applications, as fine-tuning or additional research may be needed. Although, in many case the

adjustments could be minimal. Some examples of feasible applications include:

• Art Market Analysis: Offer services to art market participants rooted in data-backed

evaluations of artworks and artists to help them to make informed decisions, identify

undervalued/ overvalued artists and to gain a better understanding market trends.

• Cultural Analytics for Galleries and Museums: Institutions could utilize our insights to

curate collections that resonate with historical significance and market demand.

• Educational Tools: Our methodology could serve as a foundation for educational programs,

offering a data-driven approach to art history.

• Authenticity Verification: Help detect misattributions by identifying inconsistencies in

image-related attributes (style, technique) and text (connections to other artists and artworks).

• Automatic Tagging and Categorization: Organize and manage large art collections

efficiently.

• Art Generation: Assist artists by suggesting new stylistic directions or provide inspiration

based on their existing work.