Format-Constraint Coupling in Knowledge Graph Construction from Statistical Tables

2026-05-21 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, medium

Summary

A study on knowledge graph construction from statistical CSV tables, specifically country-by-year time-series matrices, reveals a critical issue called "format-constraint coupling." This phenomenon describes how the serialization format and extraction schema constraints interact super-additively, with their joint effect exceeding independent effects by up to +1.180 across six datasets. Critically, applying a schema to a mismatched format can lead to catastrophic mismatch, causing fact coverage to drop below the unconstrained baseline on four of six datasets due to entity inflation or extraction refusal. Researchers attribute this to surface-form anchoring, particularly column-name references. The observed coupling holds across various LLM families and GraphRAG hosts. Furthermore, standard retrieval modes often mask these quality issues (delta <= 1pp), whereas direct graph access exposes significant gaps, up to +47.6pp (p < 0.0001). To facilitate fidelity-aware evaluation, the authors release CSVFidelity-Bench, a benchmark comprising 15 datasets, including 11 Type-II matrices and 4 Type-III tables, with 1,892 Gold Standard facts across six domains.

Key takeaway

For data scientists building knowledge graphs from statistical tables, especially time-series CSVs, you must account for "format-constraint coupling." Mismatched serialization formats and extraction schemas can severely degrade fact coverage, potentially below unconstrained baselines. Prioritize direct graph access for evaluating construction quality, as standard retrieval modes mask significant fidelity gaps. Consider using the new CSVFidelity-Bench to rigorously test your knowledge graph construction pipelines against diverse statistical table formats.

Key insights

Serialization format and schema constraints interact critically in knowledge graph construction from statistical tables, impacting fidelity.

Principles

• Format-schema mismatch can catastrophically reduce fact coverage.
• Surface-form anchoring explains format-constraint coupling.
• Direct graph access reveals quality gaps better than retrieval modes.

Method

The paper identifies "format-constraint coupling" through 2x2 factorial experiments on 6 datasets, using controlled variants across format-schema pairings, GraphRAG hosts, and LLM families. It also introduces CSVFidelity-Bench for evaluation.

In practice

• Evaluate KG construction fidelity via direct graph access.
• Use CSVFidelity-Bench for robust KG construction evaluation.
• Be aware of column-name references in schema design.

Topics

• Knowledge Graph Construction
• Statistical Tables
• Data Fidelity
• LLM Evaluation
• CSVFidelity-Bench
• Format-Constraint Coupling

Code references

• zjunlp/AdaKGC

Best for: AI Scientist, Research Scientist, Data Scientist

Related on AIssential

• Format-Constraint Coupling in Knowledge Graph Construction from Statistical Tables — Knowledge graph fidelity from statistical tables critically depends on aligning serialization format with extraction schema, as mismatches cause catastrophic failures.
• Consistency evaluation of benchmarks used for causal discovery — Benchmark causal graphs often contain mis-aligned knowledge, hindering causal discovery evaluation.
• Statistical Embeddings for Similarity, Retrieval, and Interpretable Alignment of Numeric Tabular Datasets — Statistical embeddings enable interpretable similarity and alignment of heterogeneous numeric tabular datasets for LLMs, preserving statistical context.
• EviLink: Multi-Path Schema Linking with Uncertainty-Guided Evidence Acquisition for Large-Scale Text-to-SQL — EviLink reframes Text-to-SQL schema linking as uncertainty-aware inference over multiple SQL paths, improving efficiency and accuracy.
• ScholarSum: Student-Teacher Abstractive Summarization via Knowledge Graph Reasoning and Reflective Refinement — ScholarSum integrates hierarchical knowledge graphs and iterative student-teacher refinement for fluent, factually faithful scientific summarization.
• Context Graphs: Free Association, Consistency, Narratives & Truth — LLMs function via free association on narrative graphs, trading consistency for broader, stochastically generated outputs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.