An open-source python package to identify breakthrough innovations within patents.
The Patent Breakthrough innovation (PBI) package can be used to identify breakthrough innovations in patents. The docembedding Python package contains a variety of methods for creating document embeddings. We have optimized and tested these methods for their ability to predict similarity between patents. This was done by maximizing the cosine similarity between patents that are classified into the same technology class, and minimizing cosine similarity between patents that fall into different technology classes. These methods with optimized parameters are then used to create document embeddings. From these embeddings, novelty scores are created using cosine similarities between the focal patent and patents in the previous n years and subsequent n years.