Format-Constraint Coupling in Knowledge Graph Construction from Statistical Tables

TL;DR

This paper investigates how format-constraint interactions in statistical tables affect knowledge graph extraction fidelity, revealing significant coupling effects and proposing a benchmark for fidelity-aware evaluation.

Contribution

It introduces the concept of format-constraint coupling in knowledge graph construction and provides CSVFidelity-Bench for fidelity-aware evaluation.

Findings

Format-constraint coupling exceeds additive effects by up to +1.180 in datasets.

Schema applied to mismatched formats can cause catastrophic mismatch, reducing fact coverage.

Direct graph access reveals gaps up to +47.6pp, unlike standard retrieval modes.

Abstract

An extraction schema should not reduce knowledge graph fidelity. On statistical CSV, however, it can. We study country-by-year time-series matrices, a common layout on open-data portals. In this setting, serialization format and schema constraints interact super-additively. Their joint effect exceeds the sum of independent effects by up to +1.180 (2x2 factorial, 6 datasets). Bootstrap 95% CIs are strictly positive on 4/6 datasets, with strongest evidence on wide Type-II matrices. More critically, a schema applied to a mismatched format can trigger catastrophic mismatch. Fact coverage falls below the unconstrained baseline on 4/6 datasets through entity inflation or extraction refusal. We call this observed pattern format-constraint coupling. Probing and token ablation support a surface-form anchoring explanation centred on column-name references. Controlled variants across format-schema pairings, GraphRAG hosts, and LLM families show the same direction within the measured scope; one LLM family shows only partial activation. The observation also has a diagnostic consequence. Three standard retrieval modes largely mask construction quality (delta <= 1pp), whereas direct graph access exposes gaps up to +47.6pp (p < 0.0001). To support fidelity-aware evaluation, we release CSVFidelity-Bench. It contains 15 datasets, 11 Type-II matrices, 4 Type-III tables, and 1,892 Gold Standard facts across 6 domains.