Modeling, Structuring, Storing, Querying and Appraising the Process of User Knowledge Generation in Visual Analytics

Abstract

Visual Analytics (VA) enable users to gain new knowledge through an iterative process of visualizing and interacting with data. VA's intrinsic complexity and flexibility can be used with many different goals in mind, such as data exploration or explainable AI. However, the same complexity and flexibility also impact user experience and evaluation. Domain-specific tools and complex visualizations are commonplace in VA, but they limit the number of potential users. This dissertation explores the concept of Visual Analytics Democratization wherein I seek to semi-automate the provenance of knowledge of VA workflows and model the exchange of knowledge between user and data as a knowledge graph. Such a graph can be used as a relationship database of visual insights and their underlying knowledge. It can also be used as a provenance database to relate all insights reached when using a VA tool to each other and the various steps taken for its acquisition. The proposed modeling process allows users to view and analyze the knowledge gained by past users. By linking the accumulated knowledge of ``knowledge generators,'' which include other users, AIs like ChatGPT, or knowledge bases like Wikipedia, the proposed method opens a path for democratizing the results of analysis sessions to a broader, including non-technical, audience.