Google Knowledge Graph

What the Knowledge Graph is

Google introduced the Knowledge Graph as an intelligent model that understands real-world entities and their relationships to one another, an idea the company summed up as "things, not strings" ^[1]. The contrast is with how a search engine traditionally works: a query arrives as a string of characters to be matched against documents, and the engine does not know what those characters refer to. A graph of entities takes a different route. A person, a building, a city or a film is stored as an object, and the links between objects record how they relate to each other, so the engine has something to reason over rather than only text to compare.

Where it came from

Google announced the feature in an official blog post titled "Introducing the Knowledge Graph: things, not strings" ^[1]. The Economic Information Daily reported that Google announced the feature on 16 May 2012 ^[2]. According to Google, the graph was rooted not only in public sources such as Freebase, Wikipedia and the CIA World Factbook but was also augmented at a much larger scale ^[1]. At the time of the announcement, Google said it then contained more than 500 million objects and more than 3.5 billion facts about and relationships between those objects ^[1].

What it changed on the results page

Two consequences followed from having entities rather than strings. The first was disambiguation. Google said the feature let search tell apart ambiguous terms such as Taj Mahal the monument and Taj Mahal the musician ^[1]. The same words can point at different things, and a graph of entities gives the engine a way to keep them separate and to show the one a searcher most likely meant.

The second was direct answers. Google said the information shown for Tom Cruise answers 37 percent of next queries that people ask about him ^[1]. That figure describes a feedback loop of a kind: once a summary of an entity is placed next to the results, some follow-up questions can be settled there instead of through another search. The Economic Information Daily described the same surface as a panel showing key information, giving the example of a scientist's birth and death years and scientific achievements ^[2].

How it rolled out

Google said it had begun to gradually roll out the Knowledge Graph view to US English users, and that it would also be available on smartphones and tablets ^[1]. The rollout was therefore staged rather than a single switch, and it covered phones and tablets from the start rather than desktop alone.

Where it sits in the development of search

Google described the Knowledge Graph as a first step towards a search engine that understands the world a bit more like people do ^[1]. That framing is the useful way to read the launch. Keyword matching answers the question of which documents contain the words in a query; an entity graph answers, at least in part, the question of what the query is about. The Knowledge Graph was presented as the beginning of the second approach, with hundreds of millions of objects and billions of relationships assembled as the raw material, and with public sources plus large-scale augmentation as the way that material was gathered ^[1]. Because the model is built from recorded facts and relationships, what it can show depends on what has been gathered and linked, which is why the announcement presented the work as a step rather than a finished understanding of the world.