Scaling Graph Learning for the Enterprise

Chapter 3. Traditional Machine Learning for Graphs

In this chapter, we will explore both traditional and nontraditional machine learning approaches applied to graphs. Then, we will dive into traditional graph-based machine learning, building upon the foundational concepts introduced in Chapter 2. We’ll start by exploring the nuances of graph data representation, transitioning from general methods to a focused case study on the Amazon copurchasing network. As we navigate through, we’ll uncover the diverse tasks that can be tackled using this dataset.

The heart of our exploration lies in graph feature engineering—a pivotal step that can make or break the performance of machine learning models. Here, we’ll unravel the importance of this process, the challenges encountered, and the different types of features that can be derived from graphs. Our hands-on approach will guide you through feature extraction, culminating in the integration of graph-derived metrics to enrich product attributes such as price and product category, among others. For example, we’ll demonstrate how these metrics can enhance product attributes like customer preferences, providing insights for decision making and product improvement.

Moving forward, we’ll harness these graph features to empower traditional machine learning models. You’ll gain insights into various tasks such as node classification, link prediction, and graph clustering. Through practical examples, we’ll demonstrate the process of building and ...