A generation framework of conceptual solutions integrating link prediction and large language model

In the increasingly intense market environment, both the innovativeness and the generating speed of conceptual solutions is vital for companies. In previous studies, researchers have proposed various methods for identifying innovative ideas. However, the transition from these ideas to conceptual solutions still relies on designers’ expertise and manual efforts. To address this challenge, we propose a generation framework of conceptual solutions. This framework employs semantic embedding methods to retrieve technologies relevant to requirements and utilises link prediction methods to uncover innovation opportunities. Subsequently, based on these opportunities, it leverages patent data and retrieval-augmented generation techniques to enable the large language model (LLM) to generate innovative and feasible conceptual solutions. Finally, we validated the effectiveness of the proposed framework through two case studies and demonstrated that the ConceptInnoGPT tool, developed based on this framework, significantly enhances the innovativeness, feasibility, efficiency, and variety of conceptual design. This study highlights the potential of LLM and advanced data-driven technologies in the automated generation of conceptual solutions, aiding enterprises in rapidly generating innovations to achieve a competitive edge.