Addressing Climate Change with Big Data: A Machine Learning Approach to Green Public Procurement 

Ph.D. student Bjarke Lund-Sørensen, Roskilde University, in collaboration with Professor Ole Helby Petersen and Associate Professor Lena Brogaard. 

This project examines how governments utilize the great potential to reduce greenhouse gas emissions through public procurement from the private market. Public procurement is a slightly overlooked area in the debate on climate change. However, public procurement constitutes a significant part of the economy and emissions. Public purchasing of goods, services, and construction tasks has an annual value of 14 percent of Denmark’s gross domestic product. In 2019, emissions from public procurement amounted to 12 million tons of CO2 equivalents, almost as much as the emissions from all private households. 

This calls for more knowledge on how governments use green purchasing and what influence their choice to do so. To provide a systematic answer to this, we collected all public tender and contract material in 2021-2022 for municipalities and regions. In total, we have processed 50,523 documents divided into 3.428 tenders. Due to these large volumes of documents, we have used computational resources through UCloud and HPC, including Natural Language Processing and Machine Learning to analyze the text material. 

We collect all the documents from various websites where the authorities publish them. The text material is not pre-processed and can be in many file formats. A large part of the work with HPC has therefore dealt with data cleaning and processing to ensure as uniform and meaningful data as possible. We then use the data to construct two measures for green procurement. The first is based on a relatively simple “dictionary method” based on practitioner input. We use a “green dictionary” to measure the application of green minimum requirements in the complete tenders. To construct the second measure for green tenders, we first identify the award criteria in the tenders. We then use machine learning and OpenAI’s language model to classify whether the tenders use green award criteria. HPC has been critical to both of these measures of green public procurement. 

We find an unexploited potential for green public procurement and explore several factors influencing whether local governments use green procurement, including administrative capacity, fiscal capacity, political ideology, product complexity, and asset specificity. 

The project is part of the SKILLS project. All public tender documents will be collected every week from 2021 and four years onwards. So far, we have collected over 200,000 tender documents (April 2023). 

Published date:

Associated tags: