Similarity Search in Patents via modern Artificial Intelligence

on YouTube ↗41m

" summary

Built a production Patent Management Tool that uses Python and Deep Learning to generate data vocabularies for similarity search and shows preliminary outperformance versus keyword-Boolean searches. Models train on roughly 2 billion-ish words from ~100 million patterns, use word-embedding models and libraries like spaCy and NLTK, and run in Docker with Solr and Java Tomcat. Offline batch modeling feeds an online service; performance tuning drops a costly query from 26 seconds to <1s, so the system actually works.

AI-written summary · about

" chapters (25)

1. 0:00Patent similarity search with AI
2. 1:26System overview and architecture
3. 3:55Batch and real-time layers
4. 6:04Transition to containerized architecture
5. 8:34Python-Java integration bottleneck
6. 11:43Training word models at scale
7. 12:24Data flow and NLP packages
8. 13:31Data processing performance challenges
9. 15:38Optimizing POS tagging with SpaCy
10. 17:02Phrase chunking and majority logic
11. 18:26Marking up and combining phrases
12. 20:14Next steps for processing and search
13. 20:35Visualizing word model clusters
14. 21:43Raw model experiments and entities
15. 24:12Patent filing patterns and relations
16. 26:02Legacy keyword-based search comparison
17. 27:24Key takeaways and NLP challenges
18. 28:49Q&A: Topic modeling and prior art
19. 30:39Q&A: Clustering and topological modeling
20. 33:06Q&A: User feedback integration
21. 33:58Q&A: Model validation and benchmarking
22. 35:40Q&A: Sentence clustering and extraction
23. 37:33Q&A: Phrase count and class modeling
24. 39:01Q&A: Market differentiation
25. 40:46Closing remarks

" related talks

• 41mExploring Patent Data with PythonFranta Polach · PyData Berlin 2014
• 22mDeconstructing the US Patent DatabaseVan Lindberg · PyCon 2018
• 36mAutomatic Citation generation with Natural Language ProcessingClaire Kelley, Sarah Kelley · PyData Seattle 2017
• 11mAI Based Patent Search For Chemical Compound & Drug DiscoveryBeju Rao, Aiden Raoufinia · GovAI Summit SLED 2024
• 21mData-Driven NLP: From Statistics to GenAIShahzeb Akhtar · Conf42 Machine Learning 2025
• 42mThe Machine Understands Text — Learning and Search at Micro and Macro ScaleŁ. Kuczyński · 4Developers 2021
• 32mBuilding semantic search using Deep LearningSamiur Rahman · Data By the Bay 2016
• 22mThe Capabilities of Semantic SearchFran Peric · Voxxed Days Zürich 2022