Building Knowledge Graphs with LLMs from USPTO Patent Data
summary
Fifty years of USPTO records become a Neo4j knowledge graph with stored vector embeddings and UMAP plots, OpenAI GPT simplifies abstracts into plain descriptions, RAG searches return up to 98% cosine similarity (one example shows 97% similarity) to flag likely infringement, PageRank surfaces influential patents, and a Streamlit UI lets users drop a pin, run RAG similarity, generate candidate inventions with an LLM and export to Excel; an LSTM forecasting pipeline predicts future patent clusters but outputs require subject‑matter review.
AI-written summary · about
chapters (8)
- 0:00Introduction and agenda
- 1:07Patent database use cases
- 2:58Schema design and exploration
- 5:45Simplifying legal text
- 6:50Similarity search and RAG
- 9:53Patent infringement analysis
- 12:18Predicting future innovations
- 18:01Q&A and closing
related talks
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
01h 40mKnowledge Graphs & GraphRAG: Techniques for Building Effective GenAI Applications
32mDemystifying the Mahabharata with GraphRAG
28mSearching for Research Fraud in OpenAlex with Graph Data Science
30mFast and scalable domain-specific knowledge graphs generation
31mEnhancing RAG with Fast GraphRAG and InstructLab: A Scalable, Interpretable, and Efficient Framework
37mMachine Learning models utilization
05mAugmenting Visual Information in Knowledge Graphs for Recommendations