Predicting ideas with patents

When you invent something, you likely get a patent. Usually this new idea doesn’t appear out of thin air, but rather depends on other innovations that inspired it. Excitingly, we have access to this information, and can form directed graphs between patents that cite each other.

Patents fall under categories, which are labelled by International Patent Classification (IPC) codes at various hierarchical levels. For example, the first letter from A to H represents a broad technical field, followed by digits for the class, and so on. You can bundle patents at different levels of detail and look at the frequency in which they cite each other.

Network of IPC citations at a precision of a single character (left) and two characters (right).

In reality, patents cite each other over time. So bundling them all together at once might be insightful for general patterns, but you can also look at how individual patents cite each other in time.

A large connected component of patent citations.

Essentially you can conduct all kinds of interesting analysis to identify influential categories of patents and even form some predictive capabilities over future innovations. This was work that I completed as a research assistant at the Bartlett Centre for Advanced Spatial Analysis (CASA) under the supervision of Prof. Elsa Arcaute.