Can LLMs Clean Up Your Mess? A Survey of Application-Ready Data Preparation with LLMs

TL;DR

The paper surveys the rapid emergence of data preparation powered by large language models (LLMs), reframing traditional pipelines as prompt-driven, agented workflows. It proposes a task-centric taxonomy for data cleaning, data integration, and data enrichment, detailing representative methods, their semantic capabilities, and corresponding limitations. By reviewing datasets and evaluation metrics, the work articulates comprehensive benchmarks and performance considerations, including costs, hallucinations, and cross-modal generalization. The authors outline open challenges and a forward-looking roadmap toward scalable LLM-data systems, reliable agentic designs, and robust evaluation protocols with practical impact for analytics, decision-making, and data sharing.

Abstract

Data preparation aims to denoise raw datasets, uncover cross-dataset relationships, and extract valuable insights from them, which is essential for a wide range of data-centric applications. Driven by (i) rising demands for application-ready data (e.g., for analytics, visualization, decision-making), (ii) increasingly powerful LLM techniques, and (iii) the emergence of infrastructures that facilitate flexible agent construction (e.g., using Databricks Unity Catalog), LLM-enhanced methods are rapidly becoming a transformative and potentially dominant paradigm for data preparation. By investigating hundreds of recent literature works, this paper presents a systematic review of this evolving landscape, focusing on the use of LLM techniques to prepare data for diverse downstream tasks. First, we characterize the fundamental paradigm shift, from rule-based, model-specific pipelines to prompt-driven, context-aware, and agentic preparation workflows. Next, we introduce a task-centric taxonomy that organizes the field into three major tasks: data cleaning (e.g., standardization, error processing, imputation), data integration (e.g., entity matching, schema matching), and data enrichment (e.g., data annotation, profiling). For each task, we survey representative techniques, and highlight their respective strengths (e.g., improved generalization, semantic understanding) and limitations (e.g., the prohibitive cost of scaling LLMs, persistent hallucinations even in advanced agents, the mismatch between advanced methods and weak evaluation). Moreover, we analyze commonly used datasets and evaluation metrics (the empirical part). Finally, we discuss open research challenges and outline a forward-looking roadmap that emphasizes scalable LLM-data systems, principled designs for reliable agentic workflows, and robust evaluation protocols.

Figures (9)

• Figure 1: Application-Ready Data Preparation -- Three core tasks (i.e., Data Cleaning, Integration, and Enrichment) address key sources of data inefficiency: quality issues, integration barriers, and semantic gaps.
• Figure 2: Overview of Application-Ready Data Preparation through LLM-Enhanced Methods.
• Figure 3: Example of LLM-Enhanced Data Standardization.
• Figure 4: Example of LLM-Enhanced Data Error Processing.
• Figure 5: Example of LLM-Enhanced Data Imputation.