Explanova: Automatically Discover Data Insights in N \times M Table via XAI Combined LLM Workflow

TL;DR

Explanova presents a three-stage, AutoML-like workflow that enables automatic discovery of data insights in N × M tables using a local small LLM and XAI methods. By separating Feature Preparation, Feature-to-Feature Statistics, and Feature Modeling, and by parallelizing LLM calls, the approach aims to reduce cost while extracting credible, interpretable relationships through SHAP-based explanations and a unified $\mathrm{NLL}$ metric. Key contributions include a data-type inference and transformation step, cluster-based feature augmentation, comprehensive feature-to-feature statistics across all type pairings, a parallel SHAP stability framework, and a final credibility score that gates findings. The method is demonstrated on a DSAA 5002 market_campaign dataset, showing the ability to generate detailed, readable reports and robust insight pipelines suitable for resource-constrained environments, with identified limitations and future benchmarking directions.

Abstract

Automation in data analysis has been a long-time pursuit. Current agentic LLM shows a promising solution towards it. Like DeepAnalyze, DataSage, and Datawise. They are all powerful agentic frameworks for automatic fine-grained analysis and are powered by LLM-based agentic tool calling ability. However, what about powered by a preset AutoML-like workflow? If we traverse all possible exploration, like Xn itself`s statistics, Xn1-Xn2 relationships, Xn to all other, and finally explain? Our Explanova is such an attempt: Cheaper due to a Local Small LLM.

Figures (2)

• Figure 1: Illustration of our motivation.
• Figure 2: Demostration of our designed workflow.