Exploring Patent Data with Python
summary
A patent recommendation engine for search uses Pythonic NLP and topic modeling tools Gensim + NLTK + LDA to index US patents (1975–present) totaling 70 GB compressed text. LDA models around 100 topics and finds most documents occupy 10–25 topics per document. Document similarity uses Gensim’s Similarity index to rank new abstracts; data cleaning relies on the Colman Funk Institute C++ disambiguator. Prototypes integrate Neo4j for citation graphs and Elasticsearch for full‑text, and the pipeline scales while keeping retrieval latency practical.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:55Motivation for patent data
- 5:19Patent data challenges
- 9:45Patent data workflow
- 12:41Topic modeling for patents
- 26:20Evaluating topic models
- 30:56Current experiments
- 33:54Q&A
related talks
41mSimilarity Search in Patents via modern Artificial Intelligence
36mAutomatic Citation generation with Natural Language Processing
22mDeconstructing the US Patent Database
21mData-Driven NLP: From Statistics to GenAI
48mTopic Modelling (and more) with NLP framework Gensim
35mHow does text become data?
40mOpen Data, Networks and the Law
42mThe Machine Understands Text — Learning and Search at Micro and Macro Scale