Requirement-service mapping technology in the industrial application field based on large language models

The article introduces a method of requirements-service mapping based on large-scale language models, utilizing the significant semantic understanding capability of large language models. It leverages multiple rounds of natural language question-answering to interact with users, achieve the transformation of users’ vague requirements into structured information, and eventually map to specific application services. Through combining large language models with traditional vector searching techniques, the micro-adjustment of large language models is realized for extracting and structuring requirements’ information without retraining or inputting massive data to build context. It presents classification of requirements and definition of service attributes to constrain and regulate content of user requirements, providing rules for large language models to express non-structured raw requirements into clear structured information. Upon obtaining the structured information, word embedding is further used to vectorize service information and requirements. The service mapping process is completed through vector matching algorithms, realizing the ultimate transformation from requirements to services. Finally, through industrial application service template, case studies have been conducted to analyze the accuracy of mapping under different requirement rules, thus ultimately demonstrating the effectiveness of the requirement-mapping method proposed in this article.