A Graph Entity Resolution Playbook
summary
Design graph models that break shared identifiers into nodes, standardize and validate properties, and label invalid or placeholder values by running GDS degree and excluding high-degree placeholders. Narrow candidates with weakly connected components, or if needed node similarity (Jaccard/overlap/cosine) or FastRP with KNN (set iteration weights ~2–3, use random-walk initializer). Score pairs with string measures (Levenshtein, Jaro‑Winkler), apply descriptor penalties, combine weights (example: 65% identifier, 35% text) and write results as person-group nodes or merge via apoc.refactor.mergeNodes.
AI-written summary · about
chapters (11)
- 0:00Introduction to ER
- 2:46ER Process Overview
- 3:24Designing the Data Model
- 6:38Data Standardization
- 10:21Identifying Candidates
- 12:52Node Similarity Techniques
- 15:23Using FastRP Embeddings
- 18:28Calculating Similarity
- 20:44Final Scoring Formula
- 22:42Storing ER Results
- 25:03Q&A and Summary
related talks
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
25mGraph Theory and Games: A Case Study on New York Times Connections
27mData Prep for Graphs
18mGraphScape: A Model for Automated Reasoning about Visualization Similarity and Sequencing
32mUnderstanding Complexly Connected Data using Graph Databases
41mUsing Graph-theory to empower patients
45mApplications of Graph in Fraud Detection
01h 26mBuilding an Interactive Network Graph to Understand Communities