Extracting and classifying spatial language terms from planning documents
Bibliographic record
Abstract
Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs)present new opportunities to advance text-based analyses, particularly for identify-ing and disambiguating geospatial information expressed as natural language text.This study explores the capabilities of an LLM to identify, classify, and create map-based representations of geospatial expressions found in urban planning documents.As a profession concerned with the spatial arrangement and interrelationships ofhuman activities, land uses, and environmental systems, much of the regulatoryinformation, professional expertise, and local knowledge that urban planning en-codes as text in reports and plans is geographic in nature. Interpreting and cross-referencing the geographic information in these documents is challenging and timeconsuming for both domain professionals (e.g., planners, lawyers) and communitymembers. This paper presents an automated pipeline that was developed to assessan LLM’s ability to extract geospatial expressions, determine their relevant spatialscale (micro, meso, macro), and to identify urban features referenced by a specificdevelopment proposal. Three types of planning documents were examined from amid-sized city (Kitchener, Ontario, Canada), namely a City-authored Official Plan(OP), private consultants’ development proposal reports, and transcripts from pub-lic meetings. A custom Retrieval-Augmented Generation (RAG) model was createdto process text hierarchically, extracting explicit and implicit geospatial expressionswith varying accuracy (e.g., F1-scores ranging from 0.440 to 0.800 across sources). Amulti-agent architecture, leveraging tools like Nominatim and DuckDB, was used toconvert these expressions into SQL queries and map representations. Results revealdistinct patterns in expression types and scales across document types, with OPdocuments favoring macro-scale expressions, plan justification reports emphasizingmicro-scale details, and public meeting transcripts reflecting informal micro-to-mesoreferences. While LLMs demonstrate strong potential, challenges such as factual in-consistencies, attribute errors, and context limitations highlight the need for humanoversight and improved metadata integration. This work advances the understand-ing of LLMs in geospatial analysis and highlights the potential to enhance urbanplanning processes through automated interpretation of geographic expressions intext
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.008 | 0.004 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".