IRIS: An Iterative and Integrated Framework for Verifiable Causal Discovery in the Absence of Tabular Data

2025-10-10

AI Summary

Traditional causal discovery methods rely on structured data, suffering from high collection costs, strong assumptions, and limited capacity to uncover novel causal relationships; while existing LLM-based approaches can identify known causal relations, they lack genuine causal discovery capability. Method: We propose the first end-to-end, dataset-free causal discovery framework that iteratively retrieves domain documents and extracts variables, synergistically integrating statistical causal discovery algorithms with large language models. It jointly discovers both known and novel causal relations and introduces a novel missing-variable recommendation mechanism to dynamically expand the causal graph. Contribution/Results: The framework enables real-time, verifiable causal discovery and latent variable completion directly from unstructured text. Experiments demonstrate significant improvements in coverage, novelty, and interpretability of discovered causal relations compared to state-of-the-art baselines.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Causal discovery is fundamental to scientific research, yet traditional statistical algorithms face significant challenges, including expensive data collection, redundant computation for known relations, and unrealistic assumptions. While recent LLM-based methods excel at identifying commonly known causal relations, they fail to uncover novel relations. We introduce IRIS (Iterative Retrieval and Integrated System for Real-Time Causal Discovery), a novel framework that addresses these limitations. Starting with a set of initial variables, IRIS automatically collects relevant documents, extracts variables, and uncovers causal relations. Our hybrid causal discovery method combines statistical algorithms and LLM-based methods to discover known and novel causal relations. In addition to causal discovery on initial variables, the missing variable proposal component of IRIS identifies and incorporates missing variables to expand the causal graphs. Our approach enables real-time causal discovery from only a set of initial variables without requiring pre-existing datasets.

Problem

• Discovering causal relations without pre-existing tabular data
• Combining statistical and LLM methods to find novel relations
• Automatically expanding causal graphs by identifying missing variables

Innovation

• Iterative framework combining statistical and LLM methods
• Automatically collects documents and extracts causal variables
• Identifies missing variables to expand causal graphs