Working with Data in a Connected World: the Power of Graph Data Science

A PyData Global 2021 Tutorial

Written by Dr. Clair J. Sullivan, Data Science Advocate, Neo4j

email: [email protected]

Twitter: @CJLovesData1

Last updated: 2021-10-29

Introduction

Data science and machine learning have traditionally revolved around creating models based on the assumption that individual data points are uncorrelated. However, this ignores a signal that could potentially be very strong: the relationships between data points. We will look at this data as a network graph, and explore how to unlock the potential using a graph database.

We will cover the following in this tutorial:

• An introduction to graphs and graph theory
• How working with a graph differs from columnar data, such as is available via SQL or Pandas dataframes
• A brief exploration of basic Python packages that can be used to interact with graphs
• Creation of a free Sandbox graph database
• A crash course in the Cypher query language
• Creation of a basic graph of the CORA database
• Generation of graph embeddings
• Evaluation and comparison of machine learning models based on word embeddings versus graph embeddings

This tutorial assumes no previous knowledge.

Some helfpul resources

• "Graph Algorithms: Practical Examples in Apache Spark and Neo4j" (free book)
• Create a Neo4j Sandbox
• Google Colab
• The Neo4j Cheat Sheet and Quick Reference
• Cypher Manual
• Neo4j Cypher Reference Card
• Advanced Cypher Query Tuning (video)
• Graph Data Science Library API Docs
• Bite-Sized Neo4j for Data Scientists (video series)
• My Medium Articles

If you find any errors in this tutorial, please let us know by creating an issue in this repo!