Scaling Creative Inspiration with Fine-Grained Functional Aspects of Product Ideas

Title of the Paper

Scaling Creative Inspiration with Fine-Grained Functional Aspects of Ideas

Paper Information

• Research Areas: Creativity Support Tools, Text Processing and Information Mining, Design Thinking and Innovation
• Keywords: Creativity Support Tools, Functional Representation, Design Space Exploration, Machine Learning, Information Extraction, Innovation Search, Design Inspiration, Engineering Design

Research Background and Problem Statement

• Identified Problems or Challenges:

  1. Resources such as products, patents, and scientific papers contain significant innovation potential, but their descriptions are often unstructured text, lacking the critical details needed to support creative interaction.
  2. Most existing methods offer limited expressiveness, requiring manual annotation or relying on knowledge bases with limited coverage.
  3. Extracting functional elements (purpose and mechanism) from large-scale text to break fixed patterns of creative thinking and explore the design space remains a significant challenge.

• Significance:

  1. Uncovering hidden inspiration to enhance cross-domain innovation and problem-solving capabilities is a crucial pathway for advancing technology and knowledge application.
  2. The creative potential embedded in large-scale text resources, such as patents and scientific papers, has not been fully exploited.

• Research Motivation and Related Work:

  1. Human creativity fundamentally relies on cross-domain structural matching and mechanism transfer (e.g., radar technology inspiring the microwave oven).
  2. Large organizations like NASA actively search for adaptive uses of technology, setting a strong application-oriented precedent for creative innovation.
  3. Current creativity support tools, such as the WordTree method and ontology-based abstract functional methods, face limitations in scalability and coverage.

Proposed Solution

• Proposed Method or Solution: The authors propose a scalable computational model capable of automatically extracting fine-grained functional aspects such as "purpose" and "mechanism" from large-scale text data, thereby creating inspiration-supporting design tools and functional maps.

• Innovations:

  1. Automated Extraction: Employing neural networks to automatically parse unstructured text and extract phrase-level functional units (purpose and mechanism).
  2. Functional Map Construction: Building a "functional network" across multiple products to showcase functional hierarchies and relationships, fostering design inspiration and creative exploration.
  3. Design Space Exploration Method: Supporting innovative interactions at different levels of abstraction, such as discovering adjacent problems and inspiration through purpose-mechanism maps.

• Implementation Steps/Key Techniques:

  1. Developing sequence labeling-based neural networks, including BiLSTM-CRF and GCN (Graph Convolutional Networks).
  2. Using text clustering methods (e.g., K-means) to construct conceptual nodes and applying frequent itemset rule mining to generate relationships between functional map nodes.
  3. Creating a prototype system comprising a "functional search engine" and a "design space exploration tool" to demonstrate the value of the proposed method in supporting creativity.

Research Outcomes

• Achievements:

  1. Functional Search Engine: Enables purpose- and mechanism-based search, offering more expressive query capabilities.
  2. Functional Concept Map: Automatically extracts functional relationships and constructs a design space exploration tool, helping innovators link to inspiration and potential solutions.
  3. Experiments demonstrate that the proposed method significantly enhances inspiration quality, with average precision outperforming strong baselines by 50%-60%.

• Comparison with Existing Solutions:

  1. Compared to traditional distributed representation-based text search, it provides finer-grained functional representation and higher result interpretability.
  2. Compared to existing abstraction methods (e.g., WordNet), it offers a more expressive and automated solution.

• Experimental or Evaluation Results:

  1. The mean average precision (MAP) for functional search tasks reached 87%, significantly higher than traditional baselines (40%-60%).
  2. In design space exploration tools, 62% of inspirations were rated as useful and novel.

• Limitations and Future Directions:

  1. Annotation Challenges: Crowdsourced annotation tasks are noisy; exploring weak supervision methods and better annotation approaches to distinguish purpose and mechanism is needed.
  2. Limitations of Functional Semantics: Current functional extraction is still influenced by surface lexical forms, with limited cross-domain abstraction matching capabilities.
  3. Future Directions: Further expanding the expressiveness of functional maps, exploring automatic optimization of functional nesting and abstraction levels, and network-based computation of inspiration value.

Conclusion

This study successfully integrates machine learning techniques with design thinking concepts through fine-grained functional representation, opening new pathways for designing creativity support tools. It demonstrates the feasibility of uncovering hidden inspiration in large-scale text resources.