Rethinking Data: A Data Scientist’s First Dive into GQLs
summary
Converted ad impression CSVs and census counts into a graph with nodes for location, time, impressions, period and ingested via Neo4j Aura/Desktop using Cypher; expected impressions = population×30 with sleep/nonwork adjustments, and anomalies flag when actual impressions exceed that (one city shows ~1,000,000 impressions vs 100,000 population). Also loads ~100 images from CelebA to link faces to attributes, surfaces mislabels and hour-by-hour spikes, and the visual graph exploration reliably reveals patterns that flat tables hide.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:09Speaker background
- 2:33Problem with anomaly detection
- 6:02Moving to graph databases
- 8:33Graph database structure
- 9:14Data ingestion process
- 10:21Visualizing anomalies
- 13:11Face recognition dataset demo
- 15:43Key takeaways from journey
- 17:06GraphQL tools overview
related talks
43mGraphs: Datastructures to Query
29mAll about graphs in Database
49mWarming up to Graphs
18mGraph analytics: From zero to hero
49mModern Graph Databases
49mGraph the planet: Wrangling GPU graph dataframes with GFQL
57mGraphing Your Way Through the Cosmos: Common Data Problems Solved with Graphs
55mLet Me Graph That For You