Democratizing Data Visualization: A Comprehensive Study of Visualization Recommendation Systems

The vast availability of data and the limited expertise across domains have driven the development of Visu- alization Recommendation Systems (VRSs), an emerging and challenging field of study aiming to help gen- erate insightful visualizations from data and support non-expert users in information discovery. Generally, VRS systems aim to emulate human analysts to identify relevant relationships in data and make appropriate design choices to represent these relationships with insightful charts. Among the diverse methodologies in this field, two prominent approaches stand out: Autonomous VRSs (A-VRSs) and Constraints-based VRSs (C-VRSs). Autonomous VRSs target advanced, autonomous functionality, aiming to mimic human ana- lytical capabilities by autonomously selecting and designing visualizations highlighting key relationships within a dataset. In contrast, constraints-based VRSs engage users more actively in the design process, assuming that users possess preliminary insights but lack the technical expertise to independently produce correct and insightful visualizations. Despite the significant potential of VRSs in practical applications, their progress is limited by several obstacles, including the absence of standardized datasets to train recommenda- tion algorithms, the difficulty of learning design rules, a scarcity of comprehensive evaluation strategies and benchmarks, and the complexity of defining quantitative criteria for evaluating the perceptual effectiveness of generated plots. This thesis addresses these critical gaps, contributing to four main areas: an in-depth review of VRS approaches and methodologies, the creation of high-quality datasets tailored to VRS tasks, the development of robust benchmarks and evaluation frameworks, and novel training methodologies for advancing VRS models. This thesis provides a comprehensive overview of both VRS branches, proposing contributions to face the different challenges that emerged in this field, with the ultimate goal of advanc- ing this research domain and enhancing its applicability for non-expert users, helping to democratize data visualization.