New York City real estate is a trillion-dollar market, and you’ve got your eye on a collection of prime NYC properties whose value will likely double in a couple of years. They’re not currently for sale, but with your excellent sales skills, you’re confident you can get the owners to sell at a fair price, if you can only determine who the owners are. To obtain the owners’ contact data, you’ll construct a knowledge graph from publicly available data that contains tax records, property deeds, and permits. You’ll scan the data for the target owners, analyze the datasets for possible relationships, develop a knowledge schema that can extract insights into your use case, and load the data into Neo4j to query and visualize the connections. When you’re done, you’ll have practical experience applying widely used graph tools to real-world data, and you’ll understand how different choices for your graph schema can lead to different insights.

You’ve set your sights on a collection of prime off-the-market New York City properties whose value is likely to double in the next couple of years. With your excellent sales skills, you’re confident you can get the owners to sell at a fair price, if you can determine who the owners are. You have a knowledge graph—built from tax records, property deeds, and permits—that identifies entities associated with the properties. These entities might not be the true owners, but identifying them could help you determine who the true owners are.

Your task is to transform the knowledge graph into a representation that can be processed by a machine learning model later. Using simple graph and NLP techniques, including node2vec—one of the most influential algorithms in the graph community—you’ll improve the quality of the data by removing the noise from the knowledge graph. You’ll convert the nodes in your graph into embeddings, analyze how well your embeddings represent the underlying knowledge graph, and develop insights on tuning the node2vec hyperparameters. When you’re done, you’ll have learned techniques for analyzing and visualizing embeddings and associating them to the original graph, helping you determine who the “hidden” owners are.

Create a powerful recommendation engine built from an ensemble of graph-based models that will help you tap into New York City’s real estate market by identifying groups of similar properties. You’ll start by working with transductive graph models (TransE and TransR) that are created specifically for knowledge graphs. Transductive learning takes observations from a specific set of training data and applies it to a specific set of test data. Next you’ll build an inductive model (GraphSAGE), which allows for generalized learning on new data (i.e. predictive modeling on previously unseen properties). Lastly, you’ll build the recommender system by using the k-Nearest Neighbor (kNN) algorithm to identify similar properties. When you’re done, you’ll have hands-on experience applying machine learning techniques to real-world knowledge graphs… and possibly a lucrative side hustle.