Explainability-Driven Quality Assessment for Rule-Based Systems

TL;DR

The paper addresses the challenge of improving rule quality in knowledge-based reasoning without labor-intensive data labeling. It proposes an explanation-driven framework that generates trace-based, contextual, contrastive, and counterfactual explanations to refine existing rules, integrated into Punya (the MIT App Inventor fork). Using a finance use case, it demonstrates how explanations support debugging, validation, and fair, transparent decision-making, with a lightweight reasoning architecture based on Apache Jena and a knowledge-graph model, including $DTI$-based inferences. The approach offers practical benefits for knowledge engineers and end users, and points to future work expanding explanation types and improving natural-language outputs for broader adoption.

Abstract

This paper introduces an explanation framework designed to enhance the quality of rules in knowledge-based reasoning systems based on dataset-driven insights. The traditional method for rule induction from data typically requires labor-intensive labeling and data-driven learning. This framework provides an alternative and instead allows for the data-driven refinement of existing rules: it generates explanations of rule inferences and leverages human interpretation to refine rules. It leverages four complementary explanation types: trace-based, contextual, contrastive, and counterfactual, providing diverse perspectives for debugging, validating, and ultimately refining rules. By embedding explainability into the reasoning architecture, the framework enables knowledge engineers to address inconsistencies, optimize thresholds, and ensure fairness, transparency, and interpretability in decision-making processes. Its practicality is demonstrated through a use case in finance.

Figures (3)

• Figure 1: Reasoning Architecture with Explainer Component
• Figure 2: The Explanation Component and its Inputs
• Figure 3: Android App Screenshots