CausaLens: A System for Summarizing Causal DAGs

Abstract

Causal inference aids researchers in discovering causal relationships, leading to scientific insights. Pearl's causal model uses causal DAGs to estimate causal effects, so DAG correctness is essential for reliable causal conclusions. However, for high-dimensional data, the causal DAGs are often complex beyond human verifiability. Graph summarization is a logical next step, but current methods for general-purpose graph summarization are inadequate for causal DAG summarization, as they are not designed to preserve causal information. In this demonstration, we present a system called CausaLens that summarizes a given causal DAG and balances graph simplification for better understanding and retention of essential causal information for reliable inference directly on the summary DAG. We illustrate that causal inference on the summary DAG is more robust to misspecification in the initial causal DAG compared to performing inference directly on the initial causal DAG, thereby enhancing the robustness of causal inference. We will demonstrate the utility of CausaLens for generating useful summary causal DAGs by interacting with the SIGMOD'25 participants, who will act as data analysts aiming to perform causal analysis on high dimensional datasets.