Chinese mineral exploration named entity recognition for literature mining by fusing multi-features with an enhancement domain pre-training model

• Human-computer interaction annotation to construct the first Chinese mineral exploration named entity recognition dataset. • The data enhancement model CnGeoPLM-MLM was constructed by combining the pre-trained model and the masking mechanism. • Proposed lexically enhanced entity recognition model by incorporating Chinese character structures CharFeature-Lexicon. Field geological work has accumulated a large number of mineral exploration reports, which comprehensively describe information such as the structure of ore bodies and resource distribution. However, the limitations of unstructured text processing techniques lead to inadequate knowledge extraction and data mining. Named Entity Recognition (NER) is an effective technique that extracts and mines valuable information by recognizing and processing unstructured geological text, but the specialization, polysemy, and irregularity of text in the field of mineral exploration, as well as the lack of a specific annotated corpus, make it difficult to improve the general NER method in terms of accuracy and effectiveness. Thus, this paper constructs a mineral exploration named entity recognition (MENER) model, CharFeature-Lexicon. Firstly, based on the mineral exploration geological report annotation combined with data attributes, a mineral exploration text data augmentation model CnGeoPLM-MLM is designed to construct a Chinese MENER dataset in combination with manual annotation. Secondly, based on SoftLexicon, the Chinese five-stroke and strokes of a Chinese character features are introduced to perform the MENER task on the constructed dataset. The results show that the CharFeature-Lexicon model, with an F1 value of 82.2%, is able to effectively recognize mineral resource entities.