Knowledge Graphs — Intermediate
Build graphs from unstructured text and run them in production: entity extraction, ontology evolution, hybrid graph+vector retrieval, entity resolution, and graph-retrieval evaluation.
- 1
Entity Extraction & Graph Construction
Turn unstructured text into a graph: named-entity recognition, relation extraction, LLM-based triple extraction with structured output, coreference, and dedup-on-construction.
50 minNERRelation ExtractionLLM Extraction - 2
Ontology Design & Schema Evolution
Ontology vs taxonomy, class hierarchies and constraints, controlled vocabularies, and how to evolve and version a graph schema without breaking the queries that depend on it.
50 minOntology DesignConstraintsSchema Evolution - 3
Hybrid Retrieval: Graph + Vector
Combine vector similarity with graph traversal: vector-seed-then-expand and graph-filter-then-rank patterns, local vs global GraphRAG, and assembling subgraph context for an LLM.
50 minHybrid RetrievalVector + GraphSubgraph Search - 4
Entity Resolution & Deduplication
Why duplicate entities wreck a graph, and how to fix them: record linkage, blocking, similarity scoring, match clustering, and safely merging nodes while re-pointing their edges.
50 minEntity ResolutionRecord LinkageDeduplication - 5
Evaluating Graph Retrieval
The intermediate capstone — measure graph and GraphRAG retrieval quality with gold subgraphs, subgraph recall, path correctness, and answer faithfulness, then run the tuning loop.
50 minGraph MetricsGold SubgraphsEvaluation