Fine-tuned LLM for Transportation Report Reasoning

Background

Intelligent Transportation Systems (ITS) enhance transportation by integrating sophisticated information and communication technologies into infrastructure and vehicles, forming a synergized suite of solutions that improve the capabilities of the overall transportation system. In recent years, ITS has been undergoing rapid growth and evolution due to technological advancements, heightened system integration, and an increased emphasis on urban mobility and sustainability. Figure below shows a map of summarized ITS evaluation results across US nationwide, reported by ITSJPO, Noblis and NYU research team. link

Given that many developments in ITS are relatively new, gleaning best practices and insights from peer projects becomes essential to minimize redundant efforts and improve shared knowledge. For instance, there might be a need for uniform data standards for emerging technologies, such as Connected Vehicles or e-micromobility. This needs considerations not just from a design perspective but also the synthesis of knowledge from multiple field pilots or research endeavors that have been conducted. However, varied methodologies may be employed in generating different Measures of Effectiveness (MOEs). Furthermore, such information might be available in disparate formats, such as research papers, reports, fact sheets, software tools or data visualizations. This diversity poses a challenge in establishing a consolidated platform for mutual learning, which demands a knowledge database to streamline the understanding of these ITS projects and facilitate technology transfer.

Recognizing the current gap in comprehensive knowledge surrounding ITS projects, there is a need to create a knowledge capture and deployment tracking platform that can monitor the deployment of ITS projects, analyzing their implications from performance metrics used to equity outcomes.

Project Description

We will develop a knowledge base to track the deployment of ITS projects and summarize their equity outcomes, performance measures, tools, benefits, costs, and best practices in contributing to US DOT strategic goals. The team will leverage its experience on a similar national ITSJPO project to collect, synthesize, and transfer knowledge. A web portal will be created to connect academic research and the state of the practice. Phase I of the project will start from University Transportation Center (UTC) research, especially on congestion reduction and studies that contains equity considerations in transportation. This knowledge base will especially focus on knowledge that has not been studied well in the past practices, including categorization of new measures of effectiveness (MOE) used for emerging technologies, summary of tech transfer methodology, operationalized tools and outcome of the research, as well as equity aspect of the deployments. AI-based solutions, including large language models (LLMs), will be developed to assist the knowledge synthesis process with manual validation to evaluate the effectiveness of using this AI-based approach.

Resources

Project details: link