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

Résumé

<div><p>We develop an AI system that pairs engineering problems with biology-inspired solutions at a large scale, by analyzing over 101 million abstracts to identify thematic links between engineering and biology. We detect coherent themes in each domain with transformer-based embeddings and BERTopic, then link them in a topic graph that quantifies their co-occurrence. We use TRIZ (Theory of Inventive Problem Solving) analysis to show how biological principles can overcome specific engineering limitations. By integrating language models, topic modeling, and contradiction analysis, the approach highlights latent thematic overlaps. Our methodology is demonstrated in four distinct case examples-including adhesive mechanisms for robotic climbing and thermal insulation inspired by dental bonding-validating our approach. This systematic approach can accelerate the discovery of new bio-inspired innovations.</p></div>

BibTeX

@article{douard:hal-05280378,
  TITLE = {{Large-scale transformer-based topic graphs identify thematic links between engineering and biology}},
  AUTHOR = {Douard, Nicolas and Cavallucci, Denis and Samet, Ahmed and Giakos, George},
  URL = {https://univoak.hal.science/hal-05280378},
  JOURNAL = {{Scientific Reports}},
  PUBLISHER = {{Nature Publishing Group}},
  VOLUME = {15},
  NUMBER = {1},
  PAGES = {29256},
  YEAR = {2025},
  MONTH = Aug,
  DOI = {10.1038/s41598-025-15067-9},
  KEYWORDS = {Knowledge graphs ; Bioinspired innovation ; Transformer-based semantic embeddings ; TRIZ contradictions ; Semantic topic modeling ; Engineering-biology synergy ; Interdisciplinary research ; Natural language processing (NLP) ; Natural language processing (NLP) Interdisciplinary research Engineering-biology synergy Semantic topic modeling TRIZ contradictions Transformer-based semantic embeddings Knowledge graphs Bioinspired innovation},
  PDF = {https://univoak.hal.science/hal-05280378v1/file/s41598-025-15067-9.pdf},
  HAL_ID = {hal-05280378},
  HAL_VERSION = {v1},
}