Art History Data across Languages and Regions : Bias Detection and the Evaluation of Visualization Technique

Abstract

Digital tools and platforms have been progressively used in art history research, raising a need to understand if and how biases in the very art historical data shape the representation of art history. This thesis investigates the presence of language and regional bias in open-source databases, specifically within Wikipedia’s multilingual editions, and evaluates visualization techniques for representing art historical data. By combining quantitative bias analysis with visualization evaluation through a structured user study, this research contributes to both digital art history and data visualization methodology. For the examination of bias in open-source databases, data from six Wikipedia language editions, Wikidata, and the Getty ULAN were analyzed using a set of key performance indicators (KPIs): Language Match Rate, Local Hero Score, representation ratio, and article metadata. The thesis examined the extent to which each Wikipedia language edition favored artists associated with their respective language and region. The results of the analysis confirmed that specific Wikipedia editions tended to favor artists from their region and in turns under-represent artists that were not from their respective region, and found disparities in the article completeness and quality. These findings underscore how the editorial pool shapes art historical knowledge and the high importance of cross-refencing data. The second research objective focused on the evaluation of different visualization techniques for representing the movements and distribution of artists. Heatmaps, flow maps and Sankey diagrams were thus evaluated through a user study, which included participants with and without prior art historical knowledge. The evaluation was done via quantitative and qualitative questions that analyzed the achieved accuracy and subjective perception. While the results showed that no singular visualization technique was perfect, heatmaps were the preferred option as they communicated clearly the density and geographic concentration, while flow maps were favored in visualizing movement and directionality. Through the feedback from qualitative questions, areas of improvement were highlighted and a preference for interactive and hybrid visualization. The findings of this thesis offer a dual contribution: a quantitative analysis of language and regional bias across multiple Wikipedia language editions and a comparative user-centered evaluation of visualization techniques for communicating patterns of artistic migration. The results show that there is an ongoing need for collaboration between the disciplines of computer science and art history.