<p>Chinese Word Segmentation (CWS) is a fundamental task in Chinese natural language processing (NLP). However, in the geoscience domain, the complexity and diversity of geological terminology pose significant challenges to traditional methods. Existing research indicates that issues such as ambiguous boundaries of geological terms and the scarcity of annotated data severely constrain the effectiveness of CWS in this field. To solve these problems, we propose a novel Large Language Model (LLM)-based semi-supervised framework called LLM-GeoCWS for CWS in geoscience domain. Firstly, we develop a segmentation model based on pseudo-lexicon memory mechanism and BERT to learn word boundary features and recognize complex geological terms with limited labeled data. Additionally, we design an LLM-powered self-training strategy, which uses GPT-4o to generate high-quality extra data by analyzing syntactic and lexical structure from pseudo-labels and retrieved examples, to enhance the generalization ability of the model. Our method outperformed the state-of-the-art baseline methods in extensive experiments, and ablation experiments verified the effectiveness of the proposed segmentation model and self-training strategy.</p>