Candidate Name: Hanif Ullah Sajid
Program: Public Policy
Committee Chairs: Jason Windett
Committee Members: Benjamin J. Radford, Mary Atkinson, Yaoyao Dai 
Abstract:

This dissertation aims to leverage cost-effective artificial intelligence, especially Generative AI, to track the policy activities and media agenda in Pakistan over the past 12 years, as a part of the initiation of the Comparative Agendas Project (CAP) Pakistan. In the first methodological paper, I investigate whether automatic prompt optimization improves large language models (such as OpenAI’s GPT) in understanding and classifying policy documents. In the second methodological paper, I evaluate whether parameter-efficient fine-tuning methods, specifically LoRA, an perform comparably to the current state-of-the-art, fully fine-tuned XML-RoBERTa model for policy document classification. In the third paper, I introduce Comparative Agendas Project Pakistan by outlining the data pipeline, describing the datasets,  and presenting a use case that explores the relationship between media and parliamentary agendas. Pakistan’s inclusion in the CAP will democratize our understanding of the policy patterns.