Evolving Google Search Algorithms

When I’m looking for something at a search engine, I will often start out with a particular query and then depending upon the kinds of results I see I often change the query terms I use. It appears that Google has been paying attention to this kind of search behavior from people who search like me. A patent granted to Google earlier this month watches queries performed by a searcher during a search session, and may give more weight to the words and phrases used earlier in a session like that, and might give less weight to terms that might be added on as a session continues.

This patent seems like part of an evolution of algorithms from Google that has brought us to their Hummingbird update.

An old print from the 1880s showing a cat evolving into a catcher.

There have some other patents from Google that look at information from search query sessions, and make changes to search results based upon those. I’ve written about a couple of them, and when I went to check them out, noticed that they share the same inventors, Ashutosh Garg and Kedar Dhamdhere. Both have left Google, and Ashutosh Garg now works for a company called Bloomreach, and Kedar Dhamdhere is listed at LinkedIn as working for Facebook.

In How Google May Demote Some Search Results for Subsequent Related Searches I wrote about a Google patent titled Demotion of repetitive search results. It tells us that if we perform a search for a query such as “black coats” and then follow up with a query such a “black jackets”, any results that are shared between the two might be pushed down in the results during a search for the second term, since it’s possible that we’ve seen those and didn’t find them interesting the first time we saw them.

In the post Why Google May Change Search Result Snippets, I looked at the patent Session-based dynamic search snippets, where a somewhat different approach is indicated.

When you perform two similar searches using some of the same search terms and the same result shows up, instead of Google pushing results down when the same page is shown, Google might instead re-write the snippet for the page, focusing upon the words used in the new search. This gives that page a second chance at being selected by a searcher, since it reflects both queries that a searcher performed and is relevant for both of them. This sounds like a better solution than demoting those results.

The first patent was filed in 2007. This newly granted patent was filed in 2008. The last one, where Google may change the snippet for the same result after a new query during a session surfaces it, was originally filed in 2007, and then it appears to have been refiled in 2011. It sounds like an evolution over time may have taken place in how Google might treat these search results.

The newly granted patent is:

Contextual search term evaluation
Invented by Ashutosh Garg and Kedar Dhamdhere
Assigned to Google
US Patent 8,645,409
Granted February 4, 2014
Filed: April 2, 2008

Abstract

Apparatus, systems and methods for contextual search term evaluation are disclosed. A current search query is received during a search session. A predicate subsequence in the search query is identified. A subsequent search term in the query is identified. The search term attributes of the subsequent search term are adjusted.

Search Sessions

Search sessions can be tracked by a search engine a number of ways, but usually involve a period of time where someone appears to be searching for a specific topic and performs a number of queries related to that topic. They may multi-task and search for something unrelated during that time, but if they then take up the search for queries involving the original topic again, it could be said to be in the same query session.

This notion of seeing what people search for during query sessions also springs up in patents related to synonyms and semantic search and may be very related to Google’s Hummingbird update. See: How Google May Reform Queries Based on Co-Occurrence in Query Sessions.

Predicate Queries

Someone searches for “Atlanta Weather”, and while the search results are relevant, they don’t provide exactly what the searcher was looking for. They return to Google and type in “Atlanta Weather forecast” to narrow in upon what they were looking for. Under this patent, the first terms used are referred to as a “predicate query,” and the added term is referred to as a “subsequent” query term.

The patent tells us that it might give more weight to the terms in the predicate query and less to the subsequent query term added, and may even treate the subsequent query as if it were optional.

Reading through the patent a few times, I’m not convinced that this approach really helps matters. If a searcher had started their search with the phrase “Atlanta weather forecast” they would get different results than if they started with “Atlanta weather” and then followed up with “Atlanta weather forecast.” They do tell us one reason for doing this:

However, adding new search terms to the search query may not enhance the user experience if the search engine gives equal or additional weight to the added search term.

The patent does go into a lot of detail on how they might give the subsequent query less weight in a search, but this stated reason isn’t expanded upon.

Take Aways

I prefer the approach described in the patent on changing snippets for a page when the same page shows up in the same search session for two different queries that might use some of the same words.

It’s impossible to tell which of these approaches might be in use now by Google, if any. According to LinkedIn, Ashutosh Garg left Google in 2008. I’m not sure when Kedar Dhamdhere left, but the snippet changing patent wasn’t officially filed at the USPTO until 2011, so it is possible that the updated version of that patent was worked on by only one of them or by someone else completely.

Either way, search session information is probably being used in a few different places at Google. One is in the kind of re-writing of long and complex, and often spoken queries that we see with the Hummingbird update. Another is when a knowledge panel result shows additional information about an entity described, such as Abraham Lincoln’s height being included in a knowledge panel about him, but none of the other presidents have their height listed.

Chances are good that a lot of people do ask about Lincoln’s height in queries and search sessions involving Lincoln.

What I found most interesting about this new patent is that it shows us how algorithms to address certain issues might evolve over time and change. Here is seems clear that Google was trying out different approaches with search results for similar queries during search sessions based upon earlier queries in those sessions.

I say that because the same pair of inventors are listed on the patents I’ve pointed out, and they seem to be trying out different solutions to what might be deemed the same problem.

Google’s Hummingbird is part of that evolution in that it doesn’t rely upon the previous queries of the same searcher in a search session, but instead looks at information such as historical search query sessions and what people looked at in those to help re-write someone’s first query, which may mean getting people the results they may most likely want to see for their first search.

Share

13 thoughts on “Evolving Google Search Algorithms”

  1. Fascinating stuff Bill. I think this would be useful on Facebook where you are trying to find a particular group/page/event and the right one you are looking for doesn’t show up. But if you type in the same query or change it slightly, 9 times out of 10 you get exactly the same results. In fact I find it a lot harder now to search in Facebook than ever before.

  2. This could explain why I have to spend so much time revising queries these days. Google just does not want to give up relevant results any more. It’s not quite a waste of time to use their search engine, but it’s fast approaching that point. I actually find myself using Bing more often when I need to discover new information.

  3. Thanks Bill- you offer excellent insight into what is happening in the SERPs. I am still troubled by a search engine that tries to think too far ahead of the queries being entered into it, but your slant and takeaways help me see why from a better and more balanced vantage point. You help me see what they are trying to do, better, and I thank you for it as it helps my constant approach to answering what they need. Cheers.

  4. I agree with Marty. I don’t like that the search engine ‘over thinks’ my query. I understand Google evolves and tries to keep up with needs of users. But it sometimes leads wandering off on tangents heavily unrelated to the topic I’m interested in. Doesn’t bug me enough to move to Bing, though, Michael! Thanks for the new information.

  5. Thanks for this very insightful look into Google, I appreciate the balanced view point. I for one am very curious how hummingbird and the rest of the google changes up coming are going to change the landscape of our business. One one hand, google seems to be doing better and better at supplying the results the searcher is seeking, but on the other hand it gets more and more challenging to help clients garner the rankings they want. More and more it’s not about how many keywords can you shove at google, but rather are you truly providing quality content and products / services surfers want.

  6. I agree that the method of rewriting the display snippets for relevant listings on both the predicate and subsequent searches is favorable. The results for subsequent searches should also be slightly varied from the former results of the predicate (provided the subsequent search is more defined and not some synonymous duplication of the predicate) because an intelligent searcher is redefining his/her search for more specific results. For example, searching for “giant bean bag chairs” may return ecommerce site category pages that display variations of giant bean bags, where as a subsequent search for “giant bean bag chairs for kids” should return a revised set of results (bean bag chairs specifically pertaining to children) <- the algorithm does this pretty well already. Like you Bill, in my opinion, the rewriting of display snippets should apply when a result that is relevant for both predicate and subsequent searches is displayed. This way the user may take notice of the recurring listing amongst the varied results displayed from the revised query.

  7. As always, an amazing read. I love to follow Google down the rabbit hole as they try to create a customized search engine “best friend” for every human on earth. Thank you for taking the time to tease out the intention and story behind these patents, this blog is always the first one I point to when I talk about smarter SEO.

  8. I agree that AI is very important nowadays as everyone likes a ready made material, either it be a search query suggestion or featured products on e-commerce website. But it should be till certain limit, if Google starts predicting my queries then sometimes I might not get relevant results, unless I change my query parameters few of times.

    I didn’t knew about this earlier, so I thank you for this amazing piece of information.

  9. I find it interesting how through different patents there are many ways we can see Google changing how we view their search results, I just worry that some of the changes you find through their patents could impact the way we optimise websites (as if its not hard enough as it is).

  10. Good article. I have several internet marketing friends and it just seems like such a stressful career choice to me. They constantly have to worry if today is the day google will roll out a new update and their sites will disappear. Early on with my first sites I would have considered myself a “grey hat” SEO… But now that I’m in real estate I don’t have the time to keep up with Google’s constantly changing algorithm and my site is much too valuable to my pocketbook to even toe the line of grey/black hat SEO. Truthfully, I’ve given up any sort of manipulation. Now a days it’s just quality content and high quality (relevant) back-links.

  11. There’s a lot of guess work going on here by Google. These drop down lists have the potential to kill genuine keywords. To demote something under the assumption that you didn’t find it relevant definitely has it’s flaws as you simply might not have spotted it first time around as there is usually 4-5 key phrases listed in those drop downs….or maybe you simply got distracted.

    Another thing I’ve noticed is that Google still use American spelling for Google UK, which is very frustrating when users get directed away to a spelling you may not have worked on.

  12. Hi Bill. Long time since I have commented on a post of yours (shame on me)! Another great insight here and I have to say that I am more and more troubled by Google’s algorithmic changes. on one hand they have always exuded quality (at least outwardly), but on the other hand the search is confused and disjointed, and not entirely accurate. I carry out my searches in pretty much the way you describe. I get annoyed with the apparent changes between local and national settings, never mind the actual search feedback itself.As I have been involved in the search engine business for as long as Google has been alive, I think i get more cynical about the algorithm and also wonder whether Google are deliberately trying to confuse you out of organic SERP’s and into PPC? Maybe a little off topic, but I would like your thoughts, as I have always resepcted them, and those of other contributors to this post. Cheers. Chris

  13. Very interesting stuff, Bill. This was probably the most insightful post I’ve read on how the algorithms are changing. It’s certainly food for thought. I’ve always believed (and continue to believe) that the best thing a content writer can do for SEO is to write unique, engaging content that adds value for the reader. That is by FAR the best way to become timeless in the eyes of the search engines.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>