Quality Measures for Data Visualization: A Case Study of Polish Wikipedia

This study investigates the use and distribution of data visualizations across articles in the Polish Wikipedia. Drawing on a large-scale dataset of over one million articles, we apply both algorithmic classification and direct image analysis to identify and categorize visualizations into tables, charts, diagrams, and maps. Each article was mapped to one of 22 thematic groups, enabling comparative analysis across domains such as STEM, geography, culture, and history. The dataset is enriched with metadata and semantic labels derived from a locally deployed LLaVA-Phi language model and further annotated with over 30 structural, contextual, and qualitative variables. Our results show that visualizations are unevenly distributed: tables are the most widespread across all domains, while maps dominate geographical content and charts are prevalent in technical articles. This research provides the first comprehensive overview of the visual layer of Polish Wikipedia and offers a methodological framework for evaluating encyclopedic visual content at scale. The findings contribute to ongoing discussions on visual literacy, knowledge representation, and the design of future automated systems for content enrichment and quality monitoring. Additionally, it highlights novel insights into visualization distribution patterns and identifies specific challenges related to image classification and interpretability in collaborative digital knowledge environments.