Traditional text mining research predominantly focuses on laboratory-oriented algorithms and theoretical frameworks, which hardly align with industrial practical demands. Centered on industrial implementation requirements, this paper constructs a large-model-based text mining system. It analyzes mainstream technologies applicable to industrial scenarios and designs a three-tier architecture consisting of a data layer, a model scheduling layer, and an application service layer, thereby realizing full-link integration from data governance to business services. Practical case studies verify that the proposed system can reduce costs and improve efficiency, facilitate the large-scale commercialization of natural language processing technologies, and deliver reusable engineering solutions for the digital transformation of various industries.