A retrieval-augmented method for explainable product ideation: unifying conceptual design knowledge graph and large language models

Product ideation is essential for driving innovation, especially for novel products that are newly emerging in the market and require continuous iteration to meet evolving user needs. While Large Language Models (LLMs) offer strong generative capabilities, they are prone to hallucinations due to knowledge cutoff and a lack of explainability. Conversely, knowledge graphs excel at structured design knowledge accumulation and inference, yet fall short in expressive capability to fulfill text-to-text and comprehensive design ideations. To address these limitations, this study proposes an explainable product ideation method that synergizes LLMs with an Interpretable Conceptual Design Knowledge Graph (I-CDKG) through a Retrieval-Augmented Generation (RAG) framework. The method comprises four key components: constructing a designer-friendly knowledge graph to store dynamically growing and up-to-date conceptual design knowledge; developing a graph query generation model for automatic translation of natural language design problems into queries for targeted knowledge retrieval; generating a RAG prompt using the retrieved context knowledge to augment the initial prompt fed to the LLMs; and producing grounded ideation outputs accompanied by multi-faceted explanations to help designers understand the generation logic. The method is demonstrated through a real-world design case and evaluated using both qualitative and quantitative analysis, highlighting the complementary strengths of LLMs and the I-CDKG. It is validated that the proposed method enables LLMs to generate more context-rich ideation with less hallucination while leveraging the constructed design knowledge graph to improve transparency and evidence traceability. It contributes to the development of more trustworthy and designer-centric cognitive intelligence-enabled product design.