Research on Key Technologies for Domain-specific Large Language Models in Digital Intelligence Business

With the rapid development of global digital and intelligent commerce，large language model has become a core driving force，reshaping industrial chains and value chains. This paper systematically reviews research progress on key technologies for domain-specific adaptation of large models in digital and intelligent commerce，analyzing five dimensions：computing power scheduling，algorithm optimization，data governance，architecture adaptation，and application development and operation management. The study finds that domain-specific adaptation of large models faces core challenges including：bottlenecks in distributed computing power scheduling，balancing lightweight design with performance，incomplete data governance systems，insufficient adaptation of foundational model architectures，and lack of full lifecycle management. To address these challenges，the paper organizes current research advances，including：dynamic scheduling strategies based on computing power networks，parameter-efficient fine-tuning and lightweight algorithms，multimodal data governance frameworks，modular and incremental architecture optimization，and intelligent operation systems covering the full lifecycle. Additionally，the paper explores future research directions such as：cross-regional collaborative computing power scheduling，task-specialized algorithm design，dynamic data ownership mechanisms，elastic architecture optimization，and trustworthy human-AI collaboration. From the perspective of integration between technology and management，it provides theoretical references for further advancing domain-specific adaptation of large models to promote digital and intelligent commerce and industrial innovation.