CS224W: Machine Learning with Graphs | 2021 | Lecture 3.1 - Node Embeddings

For more information about Stanford’s Artificial Intelligence professional and graduate programs, visit: Jure Leskovec Computer Science, PhD From previous lectures we see how we can use machine learning with feature engineering to make predictions on nodes, links and graphs. In this video we’ll focus on a new technique called graph representation learning that could alleviate the need to do feature engineering. In graph representation learning, we can map nodes into an embedding space such that similarity of nodes in graph is reflected by distance between their embeddings. We introduce the general components in the node embedding algorithm, namely the encoder and decoder, as well as how to define the similarity function. To follow along with the course schedule and syllabus, visit: 0:00 Introduction 0:12 Recap: Traditional ML for Graphs 1:31 Graph Representation Learning 2:40 Why Embedding? 3:26 Example Node Embedding 5:03 Setup 5:35 Embedding Nodes 7:19 Learning Node Embeddings 7:53 Two Key Components 8:47 “Shallow“ Encoding 11:27 Framework Summary 12:26 How to Define Node Similarity? 13:24 Note on Node Embeddings