Deconstructing the US Patent Database
summary
A pipeline that downloads nine million patents (soon ten million), ingests the USPTO MySQL dumps and Google BigQuery exports, computes pre-trained word vectors, maps multiword concepts to CUIs, and runs an automatic patent landscaping model in TensorFlow from a shared Jupyter notebook. The concrete claims: neural embeddings and functional ontologies find prior art, summarize portfolios, and even generate candidate inventions; the demos and datasets are public and the workflow actually works on large-scale real patents.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:01Killing Patents with Python
- 2:01The hardest thing to do
- 2:41Pre parse data
- 5:36Getting the data
- 7:55Best data
- 8:25Precanned visualizations
- 9:29Google Public Dataset
- 9:39Clean Data
- 10:12Patent Landscape
- 10:28Flowchart
- 11:49Jupiter
- 12:17Uniqueness of Patents
- 14:49What can you do
- 15:07Functional Descriptive
- 16:09Real Patent
- 16:50See You Eyes
- 18:36Concept Identifiers
- 18:57Migraines
related talks
41mExploring Patent Data with Python
41mSimilarity Search in Patents via modern Artificial Intelligence
21mData-Driven NLP: From Statistics to GenAI
11mAI Based Patent Search For Chemical Compound & Drug Discovery
36mAutomatic Citation generation with Natural Language Processing
44mThe Limits of Imagination : An Open Source journey
31mTaking AI/ML to the next level: Text!
25mSQL and TensorFlow: Enabling smart queries on a data warehouse