The Knowledge Graph and Its Importance for Intelligent Assistance

What’s a knowledge graph and why is it so important to intelligent assistant technologies? At first glance, a knowledge graph seems rather mundane. It’s just a repository of basic information describing things (entities) and their relationships.

Here’s a simple example. The world has both humans and movies in it. Humans make movies and watch movies. Humans also act in movies. A knowledge graph that describes the entities “humans” and “movies” could enable a computer to discover the link between the two. The computer would know that if there’s a movie, then the movie will most likely have one or more human actors, a director, and a producer. The knowledge graph will also point the computer to other related metadata about the movie, such as its release date, how long it runs, and any awards it may have received.

WikidataThe most well-known knowledge graph is the one underpinning Google’s recent search algorithms. Google acquired Freebase, a metadata repository, back in 2010 and used it as the foundation for Wikidata. Wikidata is a free knowledge base that currently contains over 15.6 million entries and powers Google’s Knowledge Graph.

Leveraging the information about entities and relationships in Wikidata, Google can now construct result “boxes” in response to many queries. Search “What is the Wizard of Oz,” for example, and the knowledge graph answer shows up in a box listing all the relevant data about the movie. Something similar happens if you type in “How to bake a chicken” or “How old is the Eiffel Tower?” In other words, for some questions, Google no longer relies on web crawlers to find and display related website links. Instead, like a really capable intelligent assistant, Google can leverage Wikidata and the knowledge graph to give you the exact answer you’re seeking.

When a company implements an intelligent assistant (IA) technology, one of the first tasks is to make the assistant smart about the company’s products and services. Training the IA typically entails manually populating a knowledge base of possible questions and scripted responses. But if a company can connect the IA to a dynamic knowledge graph, the IA has the ability to respond much more flexibly and accurately to questions, even those that haven’t been pre-scripted.

The concept of an Enterprise Knowledge Graph (EKG) is fairly new and made possible by machine learning and Big Data technologies, including automated text analysis and graph engines. An IA that taps into an EKG can infer the context and intent of questions, generate direct answers, make recommendations, and automatically expand its understanding as the knowledge graph adds new content.

Rovi Serves Up Personalized Entertainment Content
One real-life use case of a knowledge graph is exemplified in Nuance’s recent partnership with Rovi. Nuance leverages Rovi’s technology to add personalized entertainment services to its Dragon TV voice recognition system. Rovi has compiled a knowledge graph of entertainment content on things such as TV shows, movies, actors and actresses. The knowledge graph even contains information on the emotional quality of content, such as if it’s bittersweet or heartwarming.

Based on all this data, Rovi can combine what it knows about a viewer with information in the knowledge graph to quickly find what the viewer wants and to make relevant recommendations. The Rovi knowledge graph makes Dragon TV a powerful entertainment assistant.

Will knowledge graphs underpin the future of intelligent assistance? If the answer is yes, the next question is who will own the knowledge graphs? Will each company need to create its own, or will there be EKGs for specific verticals? The answers aren’t clear yet, but the importance of the technology seems certain.



Categories: Conversational Intelligence, Intelligent Assistants, Articles

Tags: , , ,