TabClean: Reusable LLM-Synthesized Programs for Tabular Data Cleaning

2026-06-24

AI Summary

This work addresses common data quality issues in tabular data—such as missing values, inconsistent formatting, dependency violations, unit errors, and categorical ambiguities—by proposing a training-free cleaning approach. The method leverages few-shot annotations to guide large language models (LLMs) in reasoning about data flaws and automatically synthesizing reusable Python cleaning scripts equipped with guard conditions. Central to this approach is an evidence-based guarded repair mechanism that applies deterministic transformations only when specific dirty data patterns are detected and sufficiently supported by contextual evidence. By integrating data profiling, program synthesis, and cell-level validation, the framework ensures high logical precision. Evaluated on six benchmark datasets, the method achieves higher F1 scores than state-of-the-art rule-based, learning-based, and LLM-driven baselines on five datasets, while substantially reducing computational overhead and LLM invocation costs during repeated cleaning runs.

Abstract

Reliable analytics and machine-learning pipelines depend on clean tabular data, yet production tables often contain missing values, typographical errors, inconsistent formats, violated dependencies, unit mismatches, and ambiguous categorical values. Existing cleaning systems make different trade-offs. Constraint-based systems need experts to specify rules. Learning-based systems need labels or retraining. Recent LLM-based cleaners reduce setup effort, but many call an LLM on rows, cells, or repeated workflow steps, so their cost grows with table size and with every recurring batch. We present TabClean, a model-training-free system that compiles LLM reasoning into reusable guarded cleaning programs. Given a dirty table and a small annotated development set, TabClean profiles table evidence, diagnoses repair mechanisms, synthesizes executable Python transformations, validates candidates with cell-level feedback, and commits the best program for reuse on schema-compatible batches. The key abstraction is an evidence-backed guarded repair clause. A deterministic transformation may fire only when its dirty pattern, target-negative condition, evidence support, and scope constraints are satisfied. Across six benchmarks, TabClean achieves high precision, improves F1 over representative rule-based, learning-based, and LLM-based baselines on five datasets, and substantially reduces recurring runtime and API cost by replacing repeated LLM inference with deterministic program execution.

Problem

• tabular data cleaning
• LLM-based cleaning
• reusable programs
• data quality
• schema-compatible batches

Innovation

• LLM-synthesized programs
• tabular data cleaning
• reusable cleaning programs
• guard clauses
• deterministic transformation