Large-scale transformer-based topic graphs identify thematic links between engineering and biology

Area of Science:

• Artificial Intelligence
• Biomimetics
• Computational Linguistics

Background:

• Engineering innovation often faces limitations that can be overcome by nature-inspired solutions.
• Identifying these cross-disciplinary links systematically is a significant challenge.
• Large-scale text analysis offers a potential avenue for discovering bio-inspired engineering principles.

Purpose of the Study:

• To develop an AI system for large-scale pairing of engineering problems with biology-inspired solutions.
• To identify and quantify thematic links between engineering and biology domains.
• To demonstrate the utility of this approach in accelerating bio-inspired innovation.

Main Methods:

• Analysis of over 101 million abstracts using transformer-based embeddings and BERTopic for theme detection.
• Construction of a topic graph to quantify co-occurrence of themes across disciplines.
• Application of TRIZ (Theory of Inventive Problem Solving) analysis to link biological principles with engineering limitations.

Main Results:

• Identification of coherent themes within engineering and biology using advanced NLP techniques.
• Quantification of thematic overlaps through a topic graph, revealing latent connections.
• Validation of the methodology through four diverse case studies, including robotics and materials science.

Conclusions:

• The AI system effectively identifies thematic links between engineering problems and biological solutions.
• This approach systematically highlights latent overlaps, accelerating the discovery of bio-inspired innovations.
• The integration of AI, topic modeling, and TRIZ analysis provides a powerful framework for cross-disciplinary innovation.