Graph Data Modeling For Recommendations in Neo4j

Modern data is connected. Traditional SQL, document, and key value stores are not designed to quickly query deeply connected data. This is particularly problematic for recommendations systems that rely on past relationships to make future predictions. This talk is on the Neo4j graph database, the Cypher query language, and how they can be used to score and rank connected data at query time. A short overview of graph databases and graph data modeling will provide the context needed to follow the talk, along with the costs that come with Neo4j.

Transcript

1. Graph data modeling for real-time recommendations in Neo4j ! Chris Smothers
2. Posyt Match on! • keywords • lat/lon • social graph • interest graph aLivingNotebook.com
3. DB Options Relational Graph Key Value Document
4. Why use a graph database?
5. Relationships!
8. Graph DB Options TITAN Flock DB
9. How do you make a graph? $ brew install neo4j $ neo4j start $ open http://localhost:7474
10. Cypher
11. Simple Cypher
12. Querying Relationships
13. Recommendations Starting state Returned state
14. Integrations • Ruby, Python, Go, Java, PHP, Node, … • The Cypher endpoint of the REST API supports 99% of use cases.
15. Scaling Neo4j 2.0 • ~1m traversals per second • ~100b nodes and edges • Scale up not out. Think TBs of RAM • Master/slave replication • Cache sharding… • Custom sharding (yuck). Auto sharding on they way?
16. Questions chris@pos.yt