The next frontier of ranking search results

Google unveils its new DeepRank algorithm that displays more relevant search results by understanding language the way people do.

DeepRank is explained in detail in a brand new video from Google about how search works.

DeepRank is named after the deep learning methods used by BERT and the ranking aspect of the search.

Think of DeepRank as the integration of BERT with Google Search. Although it’s not as easy as it sounds.

Here’s everything we know about DeepRank so far.

What is DeepRank?

DeepRank builds on Google’s existing machine learning and natural language processing capabilities.

Natural language processing enables searchers to enter actual questions just like they would ask a friend in a text message, for example.

Google has been working on natural language processing for almost 20 years, initially with spelling corrections and understanding synonyms.

Google’s machine learning capabilities have been in development for 10 years.

Google’s latest development in natural language processing is BERT, which further improves understanding of search algorithms for natural language queries.

Search ranking has come a long way in the last two decades, but it has not gotten to the point where algorithms can understand the intricacies of language the way humans can.


Read on below

DeepRank is the first time Google Search has had a signal that understands the relationship between human terms.

Ultimately, it was designed to make search more intuitive and to make it feel like Google understands its users.

What does DeepRank do?

DeepRank picks up the subtleties of language that are natural to humans but difficult to program.

Google’s search algorithms are used to ignore stop words and leave them out of the query.

Google DeepRank: The next frontier of ranking search resultsExample of stop words in a query that are ignored.

Over time, Google has learned that these words play an important role in communicating what people really want to say.


Read on below

With DeepRank, users can formulate search queries naturally and avoid the problem of machines not getting the subtleties.

An example of the query is shown in Google’s video: “What temperature should you preheat your oven to when you’re cooking fish?”

Without DeepRank, Google’s algorithms show some useful information, but it also gets confused.

See in the screenshot below how Google’s current search algorithm returns a page of cookies for a query about fish.

When DeepRank is tested for the query, it understands that the page is about cookies, not fish, and that it will not display the result.

Google DeepRank: The next frontier of ranking search results

Google DeepRank: The next frontier of ranking search results

Here is an example that the DeepRank team is particularly proud of that includes the query “Can you get medicine for someone’s pharmacy.”


Read on below

On the left there is a result that understands and answers the specific request (thanks to DeepRank), and on the right is a more general result about filling out recipes.

Google DeepRank: The next frontier of ranking search results

These are the types of profits required to start DeepRank.


Read on below

That brings us to the important question of When does DeepRank start?

When does Google DeepRank start?

DeepRank might sound like an easy fix, but getting BERT to play well with search isn’t easy, says Google.

Google is still in the early stages of developing DeepRank and is not about to make such a significant change in search results.

DeepRank has gone through months of testing so far, and you can see some of them firsthand on Google’s video.

Behind the scenes, there are seldom recordings of Google Search engineers carefully analyzing individual queries to determine whether DeepRank supports or affects the search results.

Every change in the search is examined carefully, no matter how big or small it is.


Read on below

However, Google engineers have limited time to work on this project.

DeepRank requires a tremendous amount of computing power even when testing. If the team working on DeepRank does not make enough progress in an acceptable time, these computer resources are reassigned to another project.

The video includes more behind-the-scenes footage of a meeting presenting progress on DeepRank to the Google launch committee.

If you’ve always wanted to see the process Google Search engineers go through to approve updates and changes, this is possibly the best look we’ve had so far.

Google’s committee approved DeepRank for an eventual launch, but there is no timeframe for launch.


Read on below

That’s all Google has shared about DeepRank so far. I expect we’ll hear a lot more about this as the launch gets closer.

See the full video below. The part about natural language processing and DeepRank starts around the 42-minute mark.