ConceptVis: Exploring and Visualizing Design Concepts With Large Language Models Using Interactive Knowledge Graph

Abstract Large language models (LLMs) are capable of generating cross-domain design knowledge, opening up new possibilities for creating a myriad of design concepts for early-stage design ideation. However, the current chat-based interface fails to represent the complexity of the design space, leading to design fixation or information overload for designers. To address this challenge, we present ConceptVis, a system that organizes and symbiotically coordinates the LLM-generated design space through an interactive knowledge graph. In ConceptVis, designers can easily visualize the structure of the design space, track the generated concepts, and explore new concepts by intuitively prompting the LLM from existing nodes. Our system aims to support users in exploring a design space in both breadth and depth to ensure the diversity and quality of the generated concepts. We conducted a user study with 24 novice designers and compared the performance of ConceptVis with that of a chat-based LLM interface for concept generation. From the user study, we observed that the system prevents users from routinely prompting the LLM and receiving similar concepts. Instead, they were encouraged to leverage popular design methods and execute them efficiently with the help of the LLM. The results illustrate that supporting users in interacting with LLMs through an interactive knowledge graph can significantly enhance the user experience and improve their performance in early-stage ideation. We also emphasize the importance of developing human-centered systems that harness the capabilities of LLMs to facilitate effective human–artificial intelligence (AI) collaborative design.