UniDM: A Unified Framework for Data Manipulation with Large Language Models

TL;DR

UniDM introduces a unified, LLM-driven framework for data manipulation in data lakes by decomposing tasks into a three-step pipeline: automatic context retrieval, context data parsing, and effective target prompt construction. This design enables cross-task generality, automatic prompt generation, and improved use of contextual evidence from data lakes, achieving state-of-the-art performance across data imputation, transformation, error detection, and competitive results in entity resolution. Through extensive ablations, the authors show that each module—context retrieval, parsing, and cloze-based prompt construction—contributes meaningfully to performance while maintaining task-agnostic applicability. The approach demonstrates strong potential for reducing manual task-specific engineering in data lake management, with practical implications for scalable, automated data cleaning and integration. Future work points to deeper domain integration, efficiency improvements, and combining LLM-based methods with traditional rule-based approaches for robust deployment.

Abstract

Designing effective data manipulation methods is a long standing problem in data lakes. Traditional methods, which rely on rules or machine learning models, require extensive human efforts on training data collection and tuning models. Recent methods apply Large Language Models (LLMs) to resolve multiple data manipulation tasks. They exhibit bright benefits in terms of performance but still require customized designs to fit each specific task. This is very costly and can not catch up with the requirements of big data lake platforms. In this paper, inspired by the cross-task generality of LLMs on NLP tasks, we pave the first step to design an automatic and general solution to tackle with data manipulation tasks. We propose UniDM, a unified framework which establishes a new paradigm to process data manipulation tasks using LLMs. UniDM formalizes a number of data manipulation tasks in a unified form and abstracts three main general steps to solve each task. We develop an automatic context retrieval to allow the LLMs to retrieve data from data lakes, potentially containing evidence and factual information. For each step, we design effective prompts to guide LLMs to produce high quality results. By our comprehensive evaluation on a variety of benchmarks, our UniDM exhibits great generality and state-of-the-art performance on a wide variety of data manipulation tasks.

Figures (6)

• Figure 1: An overview of our UniDM framework.
• Figure 2: The pipeline of UniDM by taking data imputation task as the example.
• Figure 3: Explanation example for table question answering by the UniDM.
• Figure 4: Explanation example for join discovery by the UniDM.
• Figure 5: F1-score, precision and recall on join discovery task.