What is a Knowledge Graph?
Defining knowledge graphs is particularly tricky. The definition is still in development because the technology used to create them is rapidly changing. In the past ten years, knowledge graphs have become more defined and more useful in both commercial and research applications on the web. There is actually a scientific paper devoted to defining the term with more authority and research located here. Articles and books across the world have been written to define what a knowledge graph is, and even Wikipedia redirects the search to the Ontology page.
However, there is some overlap between definitions:
“A knowledge graph acquires and integrates information into an ontology and applies a reasoner to derive new knowledge.” In the world of data, information is currency. Sometimes that information needs to be protected because of regulations and laws. However, most data remains behind locked doors and spread across many different mediums, making it difficult to do anything without referencing different locations, companies, and even paying for courses and books. Something that should be simple, such as compliance, can be difficult to understand and research.
The solution to this problem is the knowledge graph, which is designed to manage data and pull all of the world’s knowledge of something into one place. While this seems to be a pipe dream of sorts because virtualization of all information hasn’t happened yet, in an ideal world, knowledge graphs would make data universally accessible.
A Practical Example
Perhaps it is because Google favors their own products, but the easiest and most comprehensive knowledge graph is Google’s “box” that pops up when you search for a person, place, or thing. For an example, we ran a superficial search on Ada Lovelace. This was pioneered in 2012, and while it did built upon previous knowledge graphs, it used unique algorithms that are pushing change into the definition of a knowledge graph.
Application in Artificial Intelligence
Knowledge graphs are smart. They rely on ontology, which defines the semantics of a data set. Typically, an ontology supports inference and allows for implicit information to be derived from explicit data. These graphs are flexible in nature, and can be revised, extended, and changed as more data is collected. In some cases, data is fed into the AI and in others, it learns from situations that occur and adds to the collateral knowledge by itself. A properly flushed out knowledge graph can not only answer a question with the correct answer, it can trace the entire path back to the roots of how and why it knows it. In some ways, this AI is already the core of many services we see every day, like chatbots, risk analysis, and even fraud detection.