Generative Prior Transformation model for mineral resources evaluation and prediction (MineralGPT): A case study of prospective target area selection for the Xiaoshan-Xiongershan area gold polymetallic deposit

Mineral resources are an important material foundation for economic and social development. The mineral resources evaluation and prediction will provide scientific basis for the development, utilization, and protection of mineral resources. The existing traditional artificial mineral resource comprehensive evaluation methods are costly, time-consuming, and have limited data processing and analysis capabilities. However, computer-based comprehensive evaluation methods often have fixed patterns and cannot incorporate as much expert knowledge as possible into the algorithms. Additionally, the utilization rate of some multi-source heterogeneous data, especially text data, is low. Given these challenges, this study transforms expert knowledge and artificial analysis methods into priori rules and proposes a novel approach for mineral resource evaluation and prediction - the Generative Prior Transformation Model, abbreviated as MineralGPT. The MineralGPT framework is driven by the description, storage and analysis of prior knowledge to support various model algorithms such as data processing and analysis, metallogenic information extraction and prospecting prediction, content generation and optimization. Taking the optimization of the gold polymetallic mine prospecting target area in the Xiaoshan-Xiongershan area as an example, experiments on the optimization model of the prospecting target area based on term weighting in MineralGPT show that MineralGPT supported by a small amount of data are almost consistent with the expert evaluation. Compared with the large-scale language model (such as ChatGPT) that requires massive data and computing power, it has the advantages of low cost, fewer limitations and high customization. MineralGPT, which introduced the rule-based description, storage and analysis of prior knowledge, provides a new method for mineral resources evaluation and prediction, and also provides a new idea for the development of a new generation of artificial intelligence technology combining rules and learning.