Abstract: As a structured knowledge base that connects entities through semantic relationships, knowledge graphs play a significant role in precise semantic search and intelligent question answering. In the cross-border industry domain, knowledge graphs can directly support automated question answering and enterprise relationship reasoning. However, constructing industrial knowledge graphs from Chinese–Vietnamese–Thai cross-border texts still faces three major challenges: dispersed semantic focus in long texts, non-standardized domain terminology, and difficulties in multilingual entity alignment. Focusing on industrial cooperation scenarios between China and Southeast Asian countries, this paper proposes a method for automatically constructing knowledge graphs based on large language models. The domain-specific ontology is first defined, followed by three successive stages—knowledge extraction, knowledge purification, and knowledge fusion—to build a Chinese–Vietnamese–Thai cross-border industrial knowledge graph, which ultimately comprises 115,177 entities and 91,168 triples. The proposed method effectively alleviates the issue of poor-quality triples generated directly by large models and establishes a fundamental knowledge base supporting cross-border industrial collaboration. Experimental results demonstrate that the constructed knowledge graph outperforms existing methods in link prediction tasks, and preliminary question-answering applications verify its satisfactory structural integrity and practical utility.