A new patent from Google dated December 28, 2021, focuses on how to interpret queries and solve them based on entity information.
I often link to Google patents in articles because I spend a lot of time learning from them.
Patents are filed to describe new inventions and spur innovation from potential competitors. They provide enough information to exclude others in the same business from copying the intellectual property of the patent filers.
Often, we find interesting information about assumptions that the creators of patents are making about search, searchers, and the Web that can make the patents an interesting read, as well.
As always, when I share the highlights in an article like this, you are encouraged to take a look at the patent itself.
I do try to explain what the patent may cover, but don’t want to cover it in so much detail that my post may seem to be a copy of the patent to indexing programs.
You may recall when Google search engineer Paul Haahr gave a presentation at SMX 2016 on “How Google Works.”
One of the important takeaways was that Google tries to identify when entities are seen in queries submitted by searchers.
That statement leads to the question of how Google might be able to tell which entity might be referred to in a query.
Google has filed a patent where they explore that topic, and that is what this post is about.
How To Better Interpret Queries
Search has evolved to receive such search queries and return results responding to the query.
However, some search engines provide search results without understanding the search query.
For example, in response to [action movie with Tom Cruise], irrelevant search results like [Last Action Hero] and [Tom and Jerry] may be returned because a part of the search query gets included in the title of the pieces of content.
Understanding the search query can help the search engine produce more meaningful results.
How might a search engine interpret queries?
The patent points out these methods:
- Receiving a query in a search domain.
- Deciding on search terms based on the query.
- Whether a search term corresponds to an entity name.
- Looking if the entity name is from metadata associated with the search domain.
- Seeing that many entity names correspond to at least a part of the number of search terms.
- Choosing an entity type and an entity score associated with each of the numbers of corresponding entity names.
- Finding a number of entity names by removing some matching entity names based on the entity score and contextual information in the received search query.
- Performing a search in the search domain with the remaining part of the number of entity names.
- Wherein each entity named in the remaining part of the number of entity names gets searched corresponding to the associated entity type.
This method to Interpret Queries can also include:
- Receiving a voice query in a search domain.
- Choosing, many voice recognition terms based on the received voice query.
- Deciding on, for each of the numbers of voice recognition terms.
- Whether at least a part of a voice recognition term corresponds to an entity name.
- Wherein the entity name gets derived from metadata associated with the search domain and wherein an entity score gets associated with the entity name.
- Determining a feasibility score for each of the number of voice recognition terms based on the entity score.
- Ranking the number of voice recognition terms based on the determined feasibility score.
- Selecting one of the numbers of ranked voice recognition terms for executing the voice query in the search domain.
This Query Interpretation patent is located at:
Methods, systems, and media for interpreting queries
Inventors: Yongsung Kim
Assignee: Google LLC
US Patent: 11,210,289
Granted: December 28, 2021
Filed: May 5, 2017
Abstract:
Mechanisms for interpreting queries to get provided.
In some implementations, a method for analyzing queries get provided, comprising:
Receiving a search query in a search domain
Determining search terms based on the search query
Determining, for each of the search terms, whether a search term corresponds to an entity name,
Wherein the entity name gets derived from metadata associated with the search domain.
In response to determining that entity names correspond to a part of the search terms
Determining an entity type and an entity score associated with each of the corresponding entity names
Determining a remaining part of the entity names by removing at least one of the matching entity names based on the entity score and contextual information in the search query
Performing a search in the search domain with the remaining part of entity names,
Each entity named in the remaining part of entity names gets searched corresponding to the associated entity type.
The Interpret Queries Patent Conclusion
When a search engine identifies that an entity is in an article, it will try to identify specifically who the entity might be.
One Google patent I wrote about in the past explained that an entity name such as “Michael Jackson” might seem to identify just one person that most people would know. After all, he was a very well-known musician and entertainer.
But there was another well-known Michael Jackson who was nothing like that first; he was known as a Director of Homeland Security.
Google does calculate confidence scores to determine which entity might be referred to when seen in a query.
This patent tells us about how Google might determine which entity is being searched for before returning results about that entity.
Keep in mind that when someone searches for “Lincoln” (an example from another Google patent) they could mean a lincoln town car, Former President Abraham Lincoln, or the City of Lincoln, Nebraska (also many other states).
If the search engine can interpret the query correctly, and show relevant answers to a searcher, they can satisfy the searcher’s informational or situational need.
There is a lot more analysis of how this patent works in the description of the patent, but I wanted to point out why it was needed and necessary.
There is too much risk of potential confusion if the search engine didn’t try to interpret a query correctly.
More resources:
Featured Image: Prabowo96/Shutterstock
