My Fifth Post About Context at Google: Adding Context Facts to Question Answers

by Posted @ Apr 02 2018


Context Facts Tell Us How Entities are Related

Tallest US Presidents

Google provides an example of how a site can use context information in a new patent that tells us about context facts.

This post is about putting facts into context, and I feel like I need to point out other posts I have written about context at Google as I start this post. Understanding context is important. Knowing that there are other patents from Google that considered context really important adds meaning to this post. Some previous posts I have written about context include:

A recent Google Patent Application tells us that People often ask “How Tall is Barack Obama?” They point out that adding context facts to an answer for that question can make the answer even more helpful:

In certain aspects, contextual information regarding a particular fact can include ranking the fact among the other similar facts. For example, it may be helpful to know that Barack Obama is six feet and one inch tall, however, it may be more helpful to know that Barack Obama is the ninth tallest United States President. As such, the height of Barack Obama can be provided in addition to the context of how the height of Barack Obama compares to other United States Presidents.

They tell us that they might choose context facts based upon scoring those facts, and how they might make lists of things such as which are the tallest presidents.

Part of the scores for those context facts may depend in part upon such things as how often recent search queries may reference the specific entity. It can involve generating natural language text corresponding to the data about that entity.

Interest in Context Facts and Related Entities can be Learned about from Query Logs

The example they provide in the patent explores, “how the structured fact ranks within a broader group.”It can also, “provide other related entities that are ranked among the broader group.”

One of the values of this approach is that in scoring contextual facts from using information that shows up in search queries, it would require less “computational complexity” than just “exhaustively searching a knowledge base of information associated with query searches.” This means that this approach “can filter out ‘unpopular’ data that does not occur frequently in recent search queries to determine context facts efficiently and accurately.”

So the context facts that a results might show may be related to what people have been searching for about an entity, and about related entities. The patent application is:

Providing Context Facts
Pub. No.: WO/2018/052685
International Application No.: PCT/US2017/048459
Publication Date: 22.03.2018
International Filing Date: 24.08.2017
Inventors: Akash Nanavati and Andrew Huse Helmer


In an aspect, a method includes receiving lists of entities, each list (i) having an associated score, (ii) being associated with a respective context fact, and (iii) ranking a subset of the entities, and for each of the lists of entities, generating, for each entity on the list, a data structure that references (i) the entity, (ii) the context fact associated with the list, (iii) the rank of the entity for the context fact, and (iv) the score for the list. The method can also include receiving data identifying a particular entity, selecting a particular data structure that references the particular entity, and providing, for output, data indicating (i) the context fact associated with the particular data structure that references the particular entity, and (ii) the rank of the entity for the context fact associated with the particular data structure that references the particular entity.

Query Logs and Context Facts Take-Aways

This approach to scoring related facts based upon queries reminds me of how Google may be building ontologies about subjects by looking at query logs related to those, which I wrote about in SEO Moves From Keywords to Ontologies and Query Patterns. I wrote about questions in query patterns that were found in that post, such as “How tall was Barack Obama?” In the patent I am writing about today, we learn that President Obama was the Ninth tallest President. So the process of using query logs to build ontologies about a subject also seems to involve ranking facts that may be related to those subjects as well, to possibly include that information in answers, or in possible other formations, like a carousel like this:

tallest US Presidents in a carousel

The patent points out how these context facts can be combined with related entities, such as those we might see in a carousel:

Each of the related entities can include a related fact. The list corresponding to the context fact can be provided along with the related entities and the related facts corresponding to each of the related entities of the respective list. In some aspects, a predetermined number of related entities may be provided at the related entity region 1 14 of the browser interface for displaying context facts. The number of related entities that are provided can be based on the list associated with the context fact, the context fact, the total number of related entities, or any combination thereof. For example, if the list region 1 12 of the context fact includes “Tallest United States Presidents,” the related entity region 1 14 can include the four tallest United States Presidents: Abraham Lincoln, Lyndon B. Johnson, Thomas Jefferson, and Franklin D. Roosevelt and the respective related fact, or height in this instance, for each.

I have written about related entities in search results in the post Related Entity Scores in Knowledge-Based Searches. The patent I wrote about in that post talked about related entity scores and this one talks about fact scores and related entity scores. Those lead me to believe that I may have to start learning about related entities and related facts when I wrote about an entity in the future. And that Google will likely be trying to understand the relationships between entities and facts and query patterns that may be associated with those.

subscribe to our newsletter

Leave a Comment