How Google Personalizes Your Search Results

Regular visitors to Search Engine Watch will remember the article from last month, Your Email Marketing May Impact Your Rankings.

As I noted in this piece, I’m a big fan of patents. I find them helpful for confirming direction. Additionally, they help reveal the kinds of approaches Google takes to address specific problems. This is an analysis of a patent Google granted on August 27, 2015: Systems and Methods for Ranking Search Results Based on User Identification of Items of Interest.

This patent was initially filed in December of 2012. Therefore, you may realize that much of the patent has already been deployed into Google’s processes.

Using a series of calculations, the objective for this study is to understand how Google has put this patent into practice.

Patent US 2015/0242512 A1

What does this particular patent say about personalization? More importantly, how does Google calculate these things? The answer to these questions will glean some insight into other areas of the algorithm.

 First, let’s look at the patent itself. The abstract reads:

“Embodiments include a computer-implemented method that includes receiving a search query associated with a user, identifying a set of search results responsive to the search query, identifying one or more search results of the set of search results that correspond to one or more items previously identified by the user as being of interest to the user, ranking the set of search results to generate a ranked set of search results, the ranking including boosting the rankings of the one or more search results that correspond to one or more items previously identified by the user as being of interest to the user, and providing search result content for display via a graphical user interface, the search result content including the ranked set of search results.”

Essentially this says that Google is patenting a method for adjusting search results based on a user’s previous indications for liking or disliking a result.

We know this has been happening for many years, however I think it’s worth mentioning this in order to highlight the notion that implementation begins with a core ranking set and the results are augmented based on a user’s indicated preferences.

Things really start to get interesting in Section 16:

“Certain embodiments include a user selecting items that are of interest to them (e.g., places the user would like to visit in the future, articles they would like to read, etc.), and, in response to receiving a search query from the user, generating and displaying a set of search results that reflect the user’s interest in the items.”

Basically this means that a result set can be adjusted based on a user’s interest in the context of a broad idea – like location – as opposed to a specific link or webpage.

Assuming that Google knows I have a penchant for patent information, if I search for a patent for Facebook, it can produce this as the second result directly under Facebook’s own page. While it’s highly probable that this result would not appear for most, this result would likely show up for myself, along with anyone else that is interested in reading about tech patents.

Here is where things become even more fascinating for those hoping to comprehend how Google runs its calculations, as opposed to only understanding the specific ranking factors.

Section 22 states:

“In some embodiments, the user is provided an opportunity to indicate their level/degree of interest.”

Followed by this in Section 23:

“In some embodiments, factor 206 may include a predetermined factor that corresponds to the level/degree of interest. For example, high, medium and low levels of interest may correspond to factors of ‘2.0,’ ‘1.5,’ and ‘1.25,’ respectively.”

This relates to the table:


Which also corresponds with this table:

We can see that pages from the first table are omitted from the second table based on lack of relevancy to the query “Palo Alto Business.” The raw score – which is only based on the perceived relevancy of the global algorithm – is also multiplied by a factor based on specific user interest.

What is most compelling here is not the specific application to personalization – it is the revelation that Google is using multipliers against its own algorithm to favor specific sites. With this comprehension of how these aspects factor in, we can extend this understanding to other areas of SEO, such as PageRank or TrustRank. They do not have to be a factor added to the raw score. Instead, they can be used after the fact as a multiplier. 

For instance, the following statement in the patent says:

“Although the illustrated embodiment includes a single table/list that includes items of various levels/degrees of interest and corresponding factors, other embodiments may include separate tables/lists corresponding to levels/degrees of interest and/or the factors.”

What we see here is that the same multiplier can be applied to many factors. Or the multiplier can be applied after these factors are used. This may seem obvious to those who are familiar with algorithms and their designs, but for the sake of clarity, here are two examples to further explain this account:

Adjusted Score = PageRank (Links + Onsite)

This exhibits how a multiplier is used to deal with multiple variables as it applies to all. In this scenario, Google factors the site link weight plus the onsite score, and then multiplies both by the site’s PageRank. This leads to the following formula:

Adjusted Score – Linking Page’s PageRank x Link + Site PageRank x Onsite

In this example, we see the PageRank of the linking page is multiplied by the link value, while also taking into account factors such as relevance and trust, along with others. This is then added to the results of the site’s PageRank, times its own onsite score.

From a personalization standpoint, this patent is interesting enough. Mainly it illustrates what we already know to be true – a click on a link in the SERP is essentially a vote in favor of a specific resource.

And multiple clicks to the same site increases its weight and ultimately the multiplier. We can see all this in the context of what the patent includes as it states:

” […] boosting the ranking of the one or more search results comprises multiplying scores associated with the one or more search results that correspond to the one or more places previously identified by the user as being of interest to the user by the factor.”

However, this patent reminded me of the importance of understanding of how we should look at signals in general. When considering a factor like PageRank, we need to remember that it can be applied in a variety of different ways and to a variety of different elements.

This factor can be applied to the weight of a link from a specific website or to the weight of a specific page. It can also be omitted from the calculations and a different factor – such as TrustRank – can be used in its place. To make things even more complicated, a variety of factors may apply to some but not to others.

A formula could look like:

Value of A Link = Anchor + PageRank x Relevancy x (TrustRank x 0.5) + PageRank x Location on Page

Here, a very simple link value is assigned. This value is produced by the anchor score plus the PageRank, and is multiplied by the relevancy score.

This is then multiplied by one-half of the TrustRank, plus the PageRank, times the score for the link’s location. Obviously this is not the scoring value of a link, but it is an example that shows how factors can be used to play-off each other.

The important takeaway here is that this example is a reminder of why it is important to not view a factor in isolation. For example, PageRank does not need to have a single impact on the ranking score assigned to a site. It can impact the values of many of the ranking factors or none at all.

The key is to approach each ranking factor as a complex structure that is capable of having an impact on elements outside of a single value. This is used throughout this patent for personalization as well as in the global algorithm.

Now how you view Google’s algorithm and it’s hundreds of signals just became even more complicated.

Related reading

Web page access
Auditor work desk, accounting business research, financial audit, tax report
Building Construction Flat Design Vector Concept Illustration
Simple Share Buttons