Many companies provide users with access to disparate services, from search to complex interactions, all of which need a large body of general and specific knowledge represented in knowledge graphs.

Here, the authors look at knowledge graphs from Microsoft, Google, Facebook, IBM, and eBay; they contain up to two billion entities and are continuously growing. The article addresses five actively pursued main challenges related to the size of the knowledge graphs: disambiguation of automatically extracted entities, type membership for entities that can belong to different categories, capturing changes in knowledge, extracting knowledge from multiple sources, and managing such enormous knowledge graphs. Other challenges to many artificial intelligence (AI) projects involve privacy and security, as well as theoretical developments in induction, verification, and consistency.

The article takes a practical approach and presents the issue from a systems developer’s point of view; however, the problems tackled are of interest to a much larger readership. The authors argue “whether different knowledge graphs can someday share core elements.” While Wikidata may offer a technical solution, the problem is deeper: knowledge representation methods are crucial in conceptualizing reality; they are not universal; and they continuously evolve in human societies. Will Wikidata implement a universal world of categories that humans should know and accept? No. And examples from today’s systems indicate that knowledge graphs will continue to offer different ways to represent the world.