GeoKG-Ore: Knowledge Graph-Augmented Large Language Models for Intelligent Gold Exploration

Gold exploration is an evidence-intensive spatial decision problem in which geological structures, host lithologies, alteration assemblages, geochemical anomalies, geophysical responses, and remote-sensing observations must be interpreted jointly across heterogeneous spatial scales. Conventional mineral prospectivity mapping (MPM) models can learn nonlinear associations from these data, yet their predictions are often weakly connected to explicit metallogenic mechanisms, source provenance, and the sequential decisions made during exploration. Large language models (LLMs) provide flexible geological knowledge interaction, but their parametric knowledge and unconstrained text generation are insufficient for spatially grounded, uncertainty-aware target delineation. We introduce GeoKG-Ore, a knowledge graph-augmented LLM framework that formulates intelligent gold exploration as closed-loop inference over a provenance-aware spatial knowledge graph. Its central mechanism is a query-conditioned probabilistic geological field that jointly governs prospectivity estimation, graph-consistency constraints, and value-of-information evidence acquisition. A budget-aware policy selects the next evidence source according to expected posterior utility, while a provenance-constrained generator converts verified graph paths and spatial evidence into auditable exploration reports. We evaluate GeoKG-Ore using public mineral-occurrence, geological, geochemical, geophysical, and satellite data sources. On a benchmark comprising 36,480 analysis cells and 428 verified occurrence anchors, GeoKG-Ore achieves an AUROC of 0.913, an AUPRC of 0.612, and a recall of 0.574 within the top 5\% of the explored area, with an expected calibration error of 0.038. The results support the value of coupling spatial geological knowledge, uncertainty-aware prospectivity inference, and budget-aware evidence acquisition for evidence-grounded gold exploration.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11582837/v1
Primary Topic
Geochemistry and Geologic Mapping
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

GeoKG-Ore: Knowledge Graph-Augmented Large Language Models for Intelligent Gold Exploration

Zongbin Wang, Sun Zhang, Lei Li, Jianbin Yao et al.
3 links
Research Square
Geochemistry and Geologic Mapping
preprint

GeoKG-Ore: Knowledge Graph-Augmented Large Language Models for Intelligent Gold Exploration

Zongbin Wang, Sun Zhang, Lei Li, Jianbin Yao, Chunwei Wu, Cheng Liu
preprint en

Abstract

Gold exploration is an evidence-intensive spatial decision problem in which geological structures, host lithologies, alteration assemblages, geochemical anomalies, geophysical responses, and remote-sensing observations must be interpreted jointly across heterogeneous spatial scales. Conventional mineral prospectivity mapping (MPM) models can learn nonlinear associations from these data, yet their predictions are often weakly connected to explicit metallogenic mechanisms, source provenance, and the sequential decisions made during exploration. Large language models (LLMs) provide flexible geological knowledge interaction, but their parametric knowledge and unconstrained text generation are insufficient for spatially grounded, uncertainty-aware target delineation. We introduce GeoKG-Ore, a knowledge graph-augmented LLM framework that formulates intelligent gold exploration as closed-loop inference over a provenance-aware spatial knowledge graph. Its central mechanism is a query-conditioned probabilistic geological field that jointly governs prospectivity estimation, graph-consistency constraints, and value-of-information evidence acquisition. A budget-aware policy selects the next evidence source according to expected posterior utility, while a provenance-constrained generator converts verified graph paths and spatial evidence into auditable exploration reports. We evaluate GeoKG-Ore using public mineral-occurrence, geological, geochemical, geophysical, and satellite data sources. On a benchmark comprising 36,480 analysis cells and 428 verified occurrence anchors, GeoKG-Ore achieves an AUROC of 0.913, an AUPRC of 0.612, and a recall of 0.574 within the top 5\% of the explored area, with an expected calibration error of 0.038. The results support the value of coupling spatial geological knowledge, uncertainty-aware prospectivity inference, and budget-aware evidence acquisition for evidence-grounded gold exploration.

Research Square
China University of Geosciences (Beijing) (CN)
Geochemistry and Geologic Mapping
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon