https://ankky-zhao.github.io/tags/llm/LLMHugoBlox Kit (https://hugoblox.com)en-usMon, 03 Nov 2025 00:00:00 +0000https://ankky-zhao.github.io/media/icon_hu_1c0e9cb08cfb822a.pnghttps://ankky-zhao.github.io/tags/llm/https://ankky-zhao.github.io/publications/geo-class/Mon, 03 Nov 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geo-class/<p>Thousands of Google Earth Engine scripts shared online encode practical knowledge about how geospatial analyses are built, from data selection and cloud masking to index calculation and export. This knowledge is scattered across code and hard to reuse.</p> <p> <figure > <div class="flex justify-center "> <div class="w-full" > <img alt="From crowdsourced scripts to a geoprocessing knowledge graph" srcset="https://ankky-zhao.github.io/publications/geo-class/geo-class-concept_hu_a41248da9cfbb0d.webp 320w, https://ankky-zhao.github.io/publications/geo-class/geo-class-concept_hu_521aba9b388ba36d.webp 480w, https://ankky-zhao.github.io/publications/geo-class/geo-class-concept_hu_93f1b53097c20be1.webp 760w" sizes="(max-width: 480px) 100vw, (max-width: 768px) 90vw, (max-width: 1024px) 80vw, 760px" src="https://ankky-zhao.github.io/publications/geo-class/geo-class-concept_hu_a41248da9cfbb0d.webp" width="760" height="241" loading="lazy" data-zoomable /></div> </div></figure> <em>Extracting themes, functions, data sources and spatial and temporal scopes from a script into a knowledge graph.</em></p> <p>Geo-CLASS coordinates a large language model with small language models. Guided by a domain schema and a schema-aware, chain-of-thought prompt, the LLM recognises entities and relations in each script. The SLMs then map the free-text descriptions onto a geoscience knowledge base (GCMD, GeoNames, the GEE data catalogue and a functional dictionary) through semantic vectorisation, cosine similarity and best-match selection, so that synonymous descriptions from different scripts end up as the same standardised entity.</p> <p>On 237 scripts sampled from 295,943, the framework improved entity and relation recognition accuracy by up to 31.9% and 12.0% over LLM baselines including Llama-3, GPT-3.5 and GPT-4o.</p> <p>The overview figure at the top shows the full Geo-CLASS framework. Figures from Zhao et al. (2026), <em>International Journal of Geographical Information Science</em>, 40(3), 808–841, .</p>https://ankky-zhao.github.io/publications/gee-ops/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/gee-ops/<p>Mines operator call relationships from Earth Engine scripts with syntax-tree parsing and frequent-pattern analysis, forming a knowledge base that improves LLM code generation.</p> <p><strong>Citation</strong><br> Hou S., Liang J., <strong>Zhao A.</strong>, Wu H. (2025). GEE-OPs: an operator knowledge base for geospatial code generation on Google Earth Engine powered by large language models. Geo-spatial Information Science. </p> <p><em>Cover figure: Framework for constructing the GEE-OPs knowledge base. From Hou et al. (2025), Geo-spatial Information Science.</em></p>https://ankky-zhao.github.io/publications/geo-fub/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geo-fub/<p>Builds an operator-function knowledge base from Earth Engine scripts, linking operators to the functions they serve, and uses it as retrieval context for geospatial code generation with LLMs.</p> <p><strong>Citation</strong><br> Hou S., <strong>Zhao A.</strong>, Liang J., Shen Z., Wu H. (2025). Geo-FuB: constructing an operator-function knowledge base for geospatial code generation with large language models. Knowledge-Based Systems, 319, 113624. </p> <p><em>Cover figure: Workflow of the Geo-FuB framework construction. From Hou et al. (2025), Knowledge-Based Systems.</em></p>https://ankky-zhao.github.io/publications/geocode-gpt/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geocode-gpt/<p>Fine-tunes an open LLM on purpose-built geospatial code corpora for pretraining, instruction tuning and evaluation, giving a model specialised in generating geospatial code.</p> <p><strong>Citation</strong><br> Hou S., Shen Z., <strong>Zhao A.</strong>, Liang J., Gui Z., Guan X., Li R., Wu H. (2025). GeoCode-GPT: a large language model for geospatial code generation. International Journal of Applied Earth Observation and Geoinformation, 104456. </p> <p><em>Cover figure: Technical workflow of GeoCode-GPT. From Hou et al. (2025), International Journal of Applied Earth Observation and Geoinformation.</em></p>https://ankky-zhao.github.io/publications/geocogent/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geocogent/<p>An LLM-based agent that plans, retrieves geospatial knowledge, and generates and iteratively debugs code to complete geospatial programming tasks.</p> <p><strong>Citation</strong><br> Hou S., Jiao H., Liang J., Shen Z., <strong>Zhao A.</strong>, Wu H. (2025). GeoCogent: an LLM-based agent for geospatial code generation. International Journal of Geographical Information Science. </p> <p><em>Cover figure: Overall architecture of GeoCogent. From Hou et al. (2025), International Journal of Geographical Information Science.</em></p>https://ankky-zhao.github.io/publications/geocolab/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geocolab/<p>Coordinates role-specialised LLM agents in a collaborative workflow that decomposes, writes, reviews and refines geospatial code.</p> <p><strong>Citation</strong><br> Wu H., Jiao H., Hou S., Liang J., Shen Z., <strong>Zhao A.</strong>, et al. (2025). GeoColab: an LLM-based multi-agent collaborative framework for geospatial code generation. International Journal of Digital Earth, 18(2). </p> <p><em>Cover figure: Concept of the GeoColab multi-agent system. From Wu et al. (2025), International Journal of Digital Earth.</em></p>https://ankky-zhao.github.io/publications/geo-modeling-kg/Sun, 01 Jun 2025 00:00:00 +0000https://ankky-zhao.github.io/publications/geo-modeling-kg/<p>Designs a geospatial modeling knowledge graph whose entities and relations are extracted with LLMs, supporting semantic retrieval and recommendation of modeling knowledge.</p> <p><strong>Citation</strong><br> Liang J., Hou S., <strong>Zhao A.</strong>, Xu Q., Xiang L., Li R., Wu H. (2025). Design and application of a semantic-driven geospatial modeling knowledge graph based on large language models. Geo-spatial Information Science. </p> <p><em>Cover figure: Overview of the co-augmentation framework. From Liang et al. (2025), Geo-spatial Information Science.</em></p>https://ankky-zhao.github.io/publications/gpt-gee-knowledge/Sat, 01 Jun 2024 00:00:00 +0000https://ankky-zhao.github.io/publications/gpt-gee-knowledge/<p>A GPT-assisted workflow that extracts structured knowledge about geographic analysis models from Google Earth Engine scripts and reuses it to support model construction.</p> <p><strong>Citation</strong><br> Liang J., <strong>Zhao A.</strong>, Hou S., Wu H., Jin F. (2024). A GPT-enhanced framework on knowledge extraction and reuse for geographic analysis models in Google Earth Engine. International Journal of Digital Earth, 17(1). </p> <p><em>Cover figure: Prompt design and interaction with ChatGPT. From Liang et al. (2024), International Journal of Digital Earth.</em></p>