ProductMeta: An Interactive System for Metaphorical Product Design Ideation with Multimodal Large Language Models

Research Background and Issues

• What problems or challenges did the authors identify?
  The authors pointed out three main challenges faced by novice designers in the process of metaphorical product design:

  1. Difficulty in discovering source concepts with creative associations to the target product.
  2. Tendency to overlook less prominent attributes while exploring diverse mapping possibilities, coupled with cognitive load.
  3. Balancing the practical usability of the product with the complexity of metaphorical expression.

• Why is this issue important?
  Metaphorical product design plays a critical role in enhancing product innovation, user experience, and emotional engagement. Successful design works can establish profound connections with users on sensory, behavioral, and socio-cultural levels through metaphors. However, these in-depth designs are often challenging to achieve due to the complexity of the design process and a lack of experience.

• Research Motivation and Related Work
  Although extensive research has explored the application of metaphors in visual design and architecture, there remains a significant gap in the study of metaphorical design support tools in the field of product design. Traditional tools such as inspiration boards and metaphor cards are limited due to their reliance on manual connections. Emerging multimodal large language models (MLLMs) have the potential to support creative design but lack domain-specific knowledge, making it difficult to systematically support metaphorical design.

Solution

• What methods or solutions did the authors propose?
  The authors developed an interactive metaphorical product design support tool called ProductMeta. This tool leverages a large language model (GPT-4o) to analyze target and source concepts and provides systematic design support through a framework-based interface.

• What are the innovative aspects of this solution?

  1. Decomposing metaphorical design into three modular processes: source exploration, mapping exploration, and design proposal construction.
  2. Generating metaphorical sources highly associated with the target product based on Norman's three levels of emotional design theory (visceral, behavioral, reflective).
  3. Supporting multi-sensory mapping to inspire designers to explore creativity across visual, tactile, auditory, and other dimensions.
  4. Balancing functional constraints and metaphorical expression during proposal construction, offering designers more flexible generation options.

• What are the implementation steps and key technologies used?

  1. Source Exploration Module: Analyzes the target product and generates metaphor cards to display emotional associations, allowing users to manually add source concepts.
  2. Mapping Exploration Module: Uses an interactive attribute combination panel to analyze mapping possibilities between target and source, supporting multi-sensory mapping exploration.
  3. Design Proposal Construction Module: Provides constraint and attribute panels, guiding designers to adjust design outputs across dimensions such as structure, functionality, and material. The system ultimately generates diverse design proposals and supports visual presentation.

Research Outcomes

• What specific outcomes were achieved?
  Through user experiments, ProductMeta demonstrated significant advantages in helping novice designers discover creative sources, explore multi-sensory mappings, and generate design proposals. The study showed that compared to existing tools (e.g., GPT-4o-based ChatGPT), ProductMeta performed better in the following aspects:

  • Provided more diverse and contextually relevant metaphorical source suggestions.
  • Better supported multi-sensory and non-linear mapping exploration.
  • Produced higher-quality design outputs, including richer creative expressions and contextual implementations.

• What were the experimental or evaluation results?

  • User experiments revealed that designers using ProductMeta generated significantly more concepts on average compared to ChatGPT (7.21 vs. 2.43), with greater focus on the discovery and mapping exploration stages.
  • In system satisfaction surveys, ProductMeta significantly outperformed ChatGPT in supporting exploration, creative expression, and transparency.
  • Expert reviews recognized that ProductMeta's generated design proposals exhibited greater contextual relevance, metaphorical depth, and emotional connection.

• Limitations and Future Directions
  Limitations include:

  1. The system's understanding of physical and functional constraints remains superficial, leading to some generated designs lacking depth and complexity.
  2. System responsiveness is affected by MLLMs' generation latency, which may indirectly weaken the user experience.
  3. AI-generated results are sometimes overly abstract or biased, requiring designers to actively filter and adjust them.

  Future research directions:

  1. Expanding the system's application to more complex design domains (e.g., automotive, architectural design).
  2. Exploring hybrid methods that combine natural language interaction with framework-based interfaces to provide more interaction options.
  3. Introducing more intuitive interaction methods (e.g., hand-drawn input) to optimize the experience of editing design details.

Through the research on ProductMeta, the authors demonstrated the facilitation of metaphorical creative design through human-computer collaboration and provided directional insights and evaluation methods for AI-based creative support tools.