Project Description

Understanding how people perceive their living environments is important for urban planning and policy-making. Traditional approaches rely on surveys or social media, but these sources often provide limited geographical coverage and skewed demographics.

This project explores whether large language models (LLMs) can be used to simulate social perceptions in a controlled setting. Students will design and run LLM-based simulations, compare the generated outputs with selected external reference data, and examine where model-driven interpretations align with or diverge from real-world patterns. The goal is to assess the potential and limitations of LLMs as a tool for social simulation while identifying systematic biases that may arise.

Key milestones

• Build a reproducible prompting pipeline for social simulations
• Generate and analyze model outputs using selected open-source LLMs
• Compare results with curated real-world datasets
• Visualize outcomes and interpret systematic differences
• Summarize insights into a final scientific report

Student requirements:

• Python experience
• Familiarity with LLM prompting or willingness to learn
• Interest in NLP, computational social science, working collaboratively toward publishing findings in an academic workshop or conference.

Apply to this project here