Computing Professional Identity for the Economic Graph
summary
LinkedIn builds an Economic Graph from 300 million members by turning noisy profile text into numeric skill vectors using endorsements, TF‑IDF weighting, LDA topic reduction and hierarchical clustering, then classifying everyone with k‑nearest neighbors. A single large employer yields 17,000 company name variants; the system reduces ~10,000 raw skills to ~200 topics and detects anomalies (karaoke endorsements). The pipeline improves job search and recommendation quality in production.
AI-written summary · about
chapters (29)
- 0:00Introduction
- 0:54Agenda
- 1:45About me
- 2:29About the talk
- 4:22Questions
- 5:14LinkedIn Vision
- 5:34LinkedIn Mission Statement
- 6:13The Economic Graph
- 12:51UCLA
- 14:21Why
- 14:46LinkedIn guide
- 15:27User experience
- 16:26LinkedIn example
- 20:14Synonyms
- 22:01Homonyms
- 23:18LinkedIn Profile
- 25:50LinkedIn Manager
- 26:59First Scale
- 30:37Hierarchical Cluster
- 32:34Dendogram
- 33:49Knearest Neighbors
- 35:00Classification
- 35:25Skill Based Approach
- 37:01Microsoft Office
- 37:32tfidf
- 38:20Data Science vs Machine Learning
- 41:36Anomaly Detection
- 42:41Visualization
- 43:44Data Cleaning
related talks
15mLabor Market Landscape Modeling with Knowledge Graphs
08mConnecting Users Via Knowledge Graph
01h 19mIntro to GraphRAG
26mScaling LinkedIn’s search infrastructure
24mGraph-Powered Job Matcher: From SQL Tables to Neo4j RAG in Production
36mHappiness inside a job: a social network analysis
24mUnlocking AI's Hidden Connections With Graphs
01h 33mBuilding an Interactive Network Graph to Understand Communities