The New York Times uses a knowledge graph to store article metadata, which can be accessed through their API. The graph is built using Neo4j AuraDB Free and includes nodes for articles, topics, people, organizations, and locations. The data model is based on the API response, with each node having properties like title, id, url, byline, source, published_date, abstract, des_facet (topics), org_facet (organizations), per_facet (people), and geo_facet (locations). The graph can be imported into Neo4j using the `apoc.load.json` procedure, which returns the response of the API as Cypher data structures. The nodes are then created and relationships are established between them. The graph can be queried to find articles by date, title, or topics, and can also be used for recommendations based on topic overlap. Other metadata like photos, authors, and geolocation can be extracted from the article data and added to the graph. The graph can be further extended with additional entity extraction and data from other publications and sources.