LLMs Take on Local Government Contracts
Ghasemizade, Mohsen
Ghasemizade, Mohsen
Citations
Altmetric:
License
License
Abstract
Interlocal agreements (ILAs) are formal contracts between local government entities that coordinate public service delivery � from policing and fire response to water systems and education. In Iowa, Chapter 28E mandates public filing of these agreements, yielding a uniquely comprehensive archive of over 21,000 contracts spanning 1993-2020. Understanding how local governments structure collaboration is critical for accountability, transparency, and public policy analysis. Yet the sheer volume, noise, and length of these unstructured legal PDFs make systematic analysis intractable through manual review alone. To address this, we introduce the Governmental Contracts Corpus � 21,629 Iowa 28E agreements � and a five-stage computational pipeline for classifying them into four institutional forms: resource sharing, service contracts, joint operations, and new joint entities. We benchmark two OCR tools, selecting SURYA (CER = 8.7%) over GOT (CER = 20%), and evaluate two parallel classification strategies: few-shot Chain-of-Thought prompting across GPT-5.2 Pro, Gemini 3 Pro, and Llama 3.1, and contextual embeddings fed into classic machine-learning classifiers. A semi-supervised expansion strategy rounds out the pipeline by routing high-confidence predictions (?70%) to human annotators to iteratively grow the labeled corpus. Our initial evaluations reveal strong performance across both strategies. An SVM on GPT embeddings achieves 0.82 overall accuracy and 0.92 on the majority 'service contracts' class. Llama 3.1 70B with 1-shot CoT prompting achieves 0.78 accuracy and excels on 'new joint entities' with an F1 of 0.86. The semi-supervised expansion further improves overall F1 by 5%. The full corpus and codebase will be publicly released in alignment with FAIR principles, enabling reproducible research on governmental transparency.
Description
Date
1/1/2026
Student Status
Graduate Student
Journal Title
Journal ISSN
Volume Title
Type of presentation
Oral Presentation
Collections
Research Projects
Organizational Units
Journal Issue
Citation
DOI
Advisor(s)
Department
Program/Major
Computer Science
College/School
College of Engineering and Mathematical Sciences
Organization
Research Category
Social Science (Govt)