Abstract: The sources of gemmological knowledge include books, journals, courses, markets and related disciplines. A complete gemmological knowledge system is of great significance to the jewelry industry. Gem knowledge points are numerous and relatively isolated in storage, which is not conducive to practitioners and researchers to retrieve knowledge. This problem can be solved by constructing a gem knowledge base system. The graph can deal with the complex association between knowledge points, which is impossible for widely used structured database, therefore, a knowledge base in the form of a knowledge graph is selected. This paper introduces the traditional knowledge graph construction method and points out the difficulties: high cost, heavy workload, difficult technology and slightly low accuracy. It is proposed to use LLM (Large language model) to complete some tasks in knowledge graph construction to improve the cost and workload. A new knowledge graph construction idea based on LLM is conceived. Its steps include data cleaning, knowledge acquisition and knowledge refinement. According to the above ideas, a gemstone knowledge graph that can cover the gemstone knowledge of the bachelor stage is constructed, and some query scenarios are displayed. The feasibility and high efficiency of the new method are proved by our test evaluation, and the possible application direction of the graph is prospected.