Quality concerns in public procurement data

Abstract

Public procurement data, despite its potential to enhance transparency and decision-making, faces significant challenges related to quality, including discoverability, accessibility, and usability. These issues hinder effective analysis and the adoption of data-driven strategies in procurement processes. This study seeks to address these gaps by characterizing the primary challenges in public procurement data quality and exploring the role of Large Language Models (LLMs), such as ChatGPT, in overcoming them. The research employs a twofold protocol: a systematic mapping study to identify and classify existing challenges and solutions, and an exploratory experimental protocol leveraging LLMs to assess their capacity to improve data usability and accessibility. Through tailored prompts and interaction modeling, the study demonstrates how LLMs can assist users, such as government suppliers, in navigating complex procurement datasets, translating technical jargon, and enhancing data discoverability. Key findings include the identification of major quality gaps in procurement data and the validation of LLMs as effective tools for addressing these challenges. The contributions of this research lie in the development of a taxonomy for public procurement data quality, the demonstration of LLM applications in this context, and recommendations for integrating advanced AI tools into public sector workflows. These insights pave the way for future studies to further optimize procurement processes through AI-driven solutions.