| Abstract: |
Design and innovation processes primarily generate knowledge upon retrieving
and synthesising knowledge of existing artefacts. Understanding the basis of
knowledge governing these processes is essential for theoretical and practical
advances, especially with the growing inclusion of Large-Language Models
(LLMs) and their generative capabilities to support knowledge-intensive tasks.
In this research, we analyse a large, stratified sample of patented artefact
descriptions spanning the total technology space. Upon representing these
descriptions as knowledge graphs, i.e., collections of entities and
relationships, we investigate the linguistic and structural foundations
through frequency distribution and motif discovery approaches. From the
linguistic perspective, we identify the generalisable syntaxes that show how
most entities and relationships are constructed at the term level. From the
structural perspective, we discover motifs, i.e., statistically dominant
3-node and 4-node subgraph patterns, that show how entities and relationships
are combined at a local level in artefact descriptions. Upon examining the
subgraphs within these motifs, we understand that artefact descriptions
primarily capture the design hierarchy of artefacts. We also find that natural
language descriptions do not capture sufficiently precise knowledge at a local
level, which can be a limiting factor for relevant innovation research and
practice. Moreover, our findings are expected to guide LLMs in generating
knowledge pertinent to domain-specific design environments, to inform
structuring schemes for future knowledge management systems, and to advance
design and innovation theories on knowledge synthesis. |