Technology and innovation mapping: turning patents and project records into maps of innovation
Technology and innovation mapping uses AI to turn large collections of text, such as patents and project descriptions, into structured maps of technologies and innovation. The Data Lab applies it to agrifood patents worldwide and to FAO's own innovation portfolio.
- A two-dimensional taxonomy of agrifood patents, by technology and by application, built from more than 3.5 million applications
- A map of FAO's innovation portfolio by theme, technology and strategic priority
- Developed with FAO's Office of Innovation (OIN)
To discuss applying the approach to other collections, contact the Data Lab.
The challenge
Information about innovation is abundant but unstructured. Patent classifications use fixed categories that were not designed around food and agriculture, and project documents describe innovation in free text, each in its own way. Without structure, it is hard to see which technologies are emerging and where effort concentrates.
How it works
AI language models read each record, such as a patent abstract or a project description, and extract structured information about the technology and its intended use. Unsupervised machine learning then groups similar records, so the categories come from the data rather than from a fixed classification.
Application: a taxonomy of agrifood patents
A fine-tuned large language model reads the titles and abstracts of more than 3.5 million agrifood patent applications filed since 1980, drawn from the PATSTAT database. The resulting dual taxonomy classifies each patent by its underlying technology and by its intended application, from value-chain stages down to specific niches.
Read more: New insights on agricultural innovation paths based on a data-driven taxonomy of technologies (ICAS X).
Application: FAO's innovation portfolio
FAO runs hundreds of projects, each described in long, free-text documents that are hard to compare across the portfolio. An open-weight AI model, running on a virtual machine within FAO, classifies each project document by the kind of innovation it involves and how it is used, and records a written rationale for every category it assigns. Experts then rate how well each category applies, through an interactive dashboard. The result shows where innovation work concentrates, where the gaps are, and where projects could learn from each other.