Riuxian Ma

Final project for the MIT class 4.550/4.570 Computation Design Lab
Development: February-May, 2025
Instructor: Prof. Takehiko Nagakura & Dr. Daniel Tsai

Overview

This research explores how to effectively transform geospatial text data generated by LLM into meaningful visual representations.

Research Question

How to effectively transform geospatial text data generated by LLM into meaningful visual representations?

Fig 1. Information Flow Structure

Example of "Gangnam Poop": Underworlds in Seoul. Grouping data and processing spatial layout.

Fig 2. Function Tags and Chart Type Allignment

Detailed functions and descriptions from text data are sorted into spatially salient graph areas.

Fig 3. Working Framework

A framework including an LLMs Geo Agent Model for urban planner, and a generative visualization model using transformer for layout generation.

Fig 4. Detailed Generative Model Input-Output Structure

Model input-output structure with label transfer via similarity (Click the image to enlarge for details).

Fig 5. Example Visualization

Priliminary visualization examples from the developed pipeline (Click the image to enlarge for details).

Fig 6. User Feedback

User responses for post-experiment questionaires.