Unlocking Real-Time News Recommendations: The Power of Graph Relationships in Neo4j
summary
Glance builds a content knowledge graph in Neo4j that ingests over 50,000 articles daily and stores a network of 50 million nodes and 150 million relationships, using Gemini for entity extraction, Vertex AI embeddings for semantic neighborhooding, and Kafka for asynchronous ingestion. Real-time recommendations avoid vector similarity at query time by traversing explicit entity and topic edges, achieving sub-100 millisecond latency, 4x Neo4j throughput via async pipeline, focused compute on top 20% articles and >85% recommendation coverage.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker intro
- 1:52Session overview and problem setup
- 3:39Challenges with traditional methods
- 4:21Three-part solution strategy
- 5:25System architecture deep dive
- 8:16Neo4j graph construction process
- 10:25Key features powered by the graph
- 12:53Trend detection and analysis
- 13:41Engineering for scale and speed
- 14:46Key takeaways and impact
- 15:28Q&A session
related talks
39mEnhancing RAG with Neo4j Knowledge Graph
47mWhat Business Can Learn from Dating
01h 11mBuilding a Recommendation Engine with Neo4j and Python
30mPerformant GraphRAG Improving Neo4j to Drive Autonomous AI ...
26mGraph Databases will change your (freakin') life
52mEnhance your work with graph database
30mHow Graph Data Science Can Turbocharge Your Knowledge Graph
19mScalable Graph Processing with Parallel Graph AnalytiX