Tuesday, July 29, 2014

How Google Identifies Entities Using Attributes?

Google has gone a way ahead in accurately identifying entities associated with the search queries. A new Google patent describes how Google might identify entities more accurately using the attributes listed in the search queries entered by the user. Any user searching for a movie name might enter the search query as:

Actor/Actress Name + movie dialog (including other words and stop words)

OR

Actor/Actress Name + song of the movie (including other words and stop words)

Example

Movie where kate Winslet says Teach me to ride like a man!

Here:

Kate Winslet is an Entity or Attribute

and

Teach me to ride like a man! is an Attribute

Entering this query gives us the following results:



The first result from the site moviefanatic.com accurately predicts the answer as "TITANIC"



This is great but Google was able to do so by correctly matching the attributes with the entity which in this case was "Titanic". 

Similar to the above example, there can be many more queries entered by the user where he/she might be looking to find book names, places names, a particular medical condition etc. Whatever be the entity, there are several features associated with the entity. By accurately identifying the features and mapping the relationship between the entity and the attributes, Google is more likely to provide an accurate answer to a user search query. 

Here is an image depicting how Google might process a query using attribute identification:

The search query entered by the user will go through the entity identification system which will predict the main entity using the matching attributes associated with that entity. The entity data store will store the individual entities while the attribute data store will store the attributes or features that those entities might possess. 

Also See: