LLM Agent Pipeline for Explainable Data Visualization Refinement

A research article introduces an agentic AI pipeline that leverages large language models to enhance the refinement of data visualization. This system tackles the difficulties associated with the exploratory analysis of high-dimensional data, which usually requires embedding data into lower-dimensional formats such as 2D or 3D for visualization purposes. Identifying suitable algorithm configurations and hyperparameter settings to create visualizations that accurately reflect the underlying data and aid in discovering patterns is challenging. By considering visualization evaluation and hyperparameter optimization as semantic tasks, the pipeline produces detailed reports that merge quantitative metrics with descriptive insights. These reports offer practical recommendations for algorithm settings, thus connecting rigorous quantitative evaluation with qualitative human understanding. The paper can be found on arXiv under identifier 2604.15319v1 with a cross announcement type.