CausalSynth: An Interactive Web Application for Synthetic Dataset Generation and Visualization with User-Defined Causal Relationships

Abstract

Understanding and inferring causal relationships between variables is a fundamental task in visualization and visual analysis. However, it can be challenging to verify inferences of causal relationships from traditional observational data because they often lack a ground truth causal model, complicating the evaluation of visual causal inference tools. To address this challenge, we introduce CausalSynth, an interactive web application designed to generate synthetic datasets from user-defined causal relationships. CausalSynth enables users to define acyclic causal graphs via a user-friendly graphical interface, establish interrelationships between variables, and produce datasets that reflect these desired causal interactions. The application also includes built-in tools for visualizing the generated datasets, facilitating deeper insights into the user-defined causal structure and aiding the validation of the generated data. By providing a user-friendly interface for synthetic data generation and visualization based on ground truth causal models, CausalSynth helps support more meaningful evaluations of visual causal inference technologies.

Publication
Poster Proceedings of IEEE VIS