The Present and Future of AI-Enabled Knowledge Graph Generation
summary
Claims that LLMs can turn text into graphs but hit an effective context window of a few thousand tokens (despite Gemini 1.5 Pro advertising 1M), and that cross‑document generation without scaffolding produces many duplicates. Uses LangChain + LLM Graph Transformer style clustering with text embeddings and reranking; Neo4j’s transformer reports ~57% node accuracy on a hard benchmark. Built a private benchmark called KG Storage and a preview system UniStorage showing much higher accuracy; projects human‑level graph generation in 6–24 months and cost falls from 19x down toward parity.
AI-written summary · about
chapters (12)
- 0:00Introduction to knowledge
- 1:09The semantic web legacy
- 2:14Use cases for graphs
- 4:16Challenges of manual markup
- 5:21LLMs for data structuring
- 7:56Overcoming model limitations
- 10:14New frontiers in research
- 12:42System performance analysis
- 15:17Achieving human-level performance
- 18:22Agentic graph generation
- 18:56Projections and the future
- 27:09Conclusion and Q&A
related talks
41mEnhancing LLMs with Graph Technology
46mKnowledge Graphs + LLMs: What's the latest in research?
44mGive Your LLMs a Left Brain
23mKnowledge Graphs and LLMs in Action
52mGive Your LLMs a Left Brain
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs