Is google's algorithm stretching the meaning of "synonym"?

[aristotle]

Recently there's been considerable discussion here about google's algorithm using synonyms as part of its process for obtaining search results.
But as I posted in another thread, I came across an example that stretches the meaning of "synonym" too far. Here is part of my other post:

I was just searching for information about a particular nutritional supplement, but the answer box and most of the first page results pertained to a different supplement. This other supplement is sometimes used for similar purposes, but it's derived from a different herb that grows in a different part of the world. It's NOT a synonym.

So even the answer box was about a different supplement.

From my other recent searches, it's clear to me that google's results have taken a turn for the worse, at least partly because they're become less relevant. In my opinion this use of synonyms has gone too far.

[engine]

[merriam-webster.com...]

Well, it is a synonym, but it's only accurate if you are looking for the other example you've cited.

I do agree, it's trying too hard, and the beta answerboxes need more work. I'm calling them beta.

[justpassing]

You can use quotes around words, in the search query.

The use of synonyms or related words is because most of people are not really sure of what they are looking for or can make mistakes, especial on special nouns/words. It happens to myself, sometimes I search for something using vaguely related words because i can't think of anything better. And Google finds/guesses what I am searching for. So I guess it's the general trend.

Keep in mind that, in the old times, this was highly educated people or / and geeks using the Internet, but now it's the average joe...

[edited by: justpassing at 8:55 pm (utc) on Nov 5, 2018]

[aristotle]

Well, it is a synonym

Hmm ... I don't know how you can say that, since you don't know what supplement I searched for.

In fact you would have to stretch the definition of "synonym" far beyond its usual meaning in order to justify what the algorithm did in this case.

In addition, that set of results was bad and of no use to me. If I had wanted information about the other supplement, I would have searched for it by its name.

[buckworks]

I agree that Google sometimes over-reaches badly when it come to synonyms. I've seen it in the search results, and I've seen it in where Google shows ads. The latter can become very expensive if you're not obsessive about restricting where your ads appear.

[MrSavage]

It's not Google's fault now. It's the AI. Are ads treated the same and if not, why not? I'm not sure if the confusion on this subject revolves around the fact that the AI pushes people towards things that everybody searches for rather than what you actually searched for. One AI, one brain for all. Sure you might have asked about X but most people search for Y, therefore Google shows you Y. I wonder if search quality is done by AI reviewing AI.

[engine]

@aristotle
In your first post you quoted

This other supplement is sometimes used for similar purposes

Hmm ... I don't know how you can say that, since you don't know what supplement I searched for.

You're right, I don't know, but you indicated it's sometimes used for similar purposes.

I'm not defending Google's result to that query, i'm just trying to get a picture as to why it had connected it, even if incorrectly. And that's the danger, incorrectly connecting.

The human brain can "understand" in ways that AI significantly lacks.

I've seen so many answer boxes that are not the answer i'm seeking, but it's google's "best guess" based on the optimisation and ranking of the site in question. Unfortunately, many people will believe the answer they get, or the answer read out on their speaker.
When users had ten blue links they did their own research from the top results. Now, they are fed the result google wants them to see.

[Shaddows]

Obviously, we are all whitehat here (can't remember the last time I saw mention of coloured hats, but never mind). But if you were looking for tricks to exploit, a careful study of not-quite-synonyms and not-quite-answer boxes would probably be a productive way to spend a week.

[RedBar]

Ooooops, a white hat just let a cat out of the box!

Try justifying that to your boss, if you have one.

[aristotle]

Well google needs to recognize that this is making their search results worse and then make adjustments to correct the problem.

[justpassing]

Not to forget that even if hundreds, thousands of even millions search results are bad, how much does it represent over the billions of searches?

[iamlost]

G has been stretching connections for a very long time all over their data usage.

However, please, please, pretty please with pink icing and a cherry on top, refrain from using the acronym 'AI' or term 'artificial intelligence' as it is but marketing spin/hype on what may (or may not) exist in years hence but is currently algorithms with one or more machine learning feedback loops. And they, including Google's, are rudimentary and buggy and barely, if that, fit for purpose.

As to Aristotle's supplement the logic train is fairly straightforward while illustrating the foibles and limitations of current machine learning: two different supplements have an overlapping use and that connection, in the current algo weightings, barring additional limiting inputs is sufficient to make them not only interchange but for the non query referenced supplement to be shown in results as or instead of the referenced one.

Google's query results go through these sorts of increasingly wrong results before correcting on a fairly regular basis. Often they only affect one vertical so the effects are not as obvious, however since the general rollout of 'RankBrain' I've noticed that the dips of 'wrong' have been deeper and longer and the corrections shallower. Overall increasing GIGO.

As justpassing mentions the likely percentage of noticeable 'errors' relative to total searches has little to no effect aka change among users. People generally tend to be oblivious except within their own narrow area of expertise.

In the beginning Google needed organic results for search, over time they have moved away from that dependency in many verticals such that often ye olde organic is is no longer 'the' product but simply filler and fallback. This was understood and talked about a decade or more ago. Often webdevs are as oblivious as the general public.

[engine]

iamlost FYI, Google's RankBrain, AI in use. Machine Learning, if you prefer.
[webmasterworld.com...]

[martinibuster]

RankBrain is so 2015.
Danny Sullivan said that Neural Matching is used in 30% of search queries. He/Google coined the phrase Super Synonyms to give a thumbnail idea of what it's doing.

[explorador]

Synonyms are not equal to alternative products, one is a pure language matter, the other comes from a bad interpretation from someone turning it into an advice (one not being asked) and also: it's a business matter.

You can use quotes around words, in the search query.

Sure, but here there are delicate implications on SEO, results and business. It could turn into G redirecting traffic away FROM YOU, or diff traffic TO YOU, as something you can't turn into sales.

[explorador]

@iamlost, I agree, a lot of people using AI referring to Artificial intelligence when it's just a series of "if" or just an algorithm. IA is something more complex. It makes me laugh seeing bad software developers referring to AI, in the past they abused the term API, now is AI.

[iamlost]

@engine: I definitely prefer ML. RankBrain then and whatever since were/are not AI in action. However, the marketing hype has transposed the two such that it should be interesting should actual AI arrive to find itself misapprehended...

@martinibuster: 'so 2015' - which is when the wheels really came off rather than merely wobbling and why I used the word 'since' so as to delineate a start without specifying an end.

As was briefly alluded earlier by Shaddows there are egregious ML mistakes in Google's query results that can be understood and leveraged - simple human intelligence advantage - to good effect. No 'bad' behaviour or ToS contravention required. The current cracks for insertion are as large as I saw prior to Panda (SEO) or 'impressions' (AdSense). And those were caverns.

[aristotle]

Machine Learning

Well the machine hasn't learned much, and what it has learned is mostly wrong. It has caused the quality of Google's search results to get much worse.

[brotherhood of LAN]

I've skimmed over the thread a few times so 'hands up' if someone has said similar. I've seen numerous occasions where I'm getting synonyms of homonyms, where I'd agree on the 'stretching of meaning'. Seems to happen more often when Google is lacking an answer so is probably it's 'next best guess'.

[lucy24]

synonyms of homonyms

That's understandable when you consider how many people are talking to smartphones instead of typing into computers--and how many more are using voice-recognition on machines that actually have a keyboard. It implies that all queries are fed into a single global let's-figure-out-what-they-meant filter, instead of going through different filters depending on whether your starting point was a written word or a spoken one--which the search engine may actually not know.

[EditorialGuy]

Not to forget that even if hundreds, thousands of even millions search results are bad, how much does it represent over the billions of searches?

Also, some errors are worse than others. If you're searching for "powdered broccoli" and you get results for "powdered brussels sprouts," that may be annoying, but it isn't as bad as an answer box that gives you the wrong advice on what to do if you think you're having a heart attack.

Another thing: Is the "bad synonym" example described here consistent in repeated searches, or could it simply be an example of testing or machine learning in the wild?

[tangor]

Well the machine hasn't learned much, and what it has learned is mostly wrong. It has caused the quality of Google's search results to get much worse.

My chuckle for the dismal is that "AI" is nothing more than bad human coding for a filter bubble that has a recursive applied. There ain't not intelligence involved, only a larger Thesaurus.

If you want to find anything, try Bing.

[martinibuster]

Bing is most definitely far behind Google now.

Google's Neural Matching algorithm is doing a remarkable job at interpreting and answering ambiguous queries. A side by side comparison between Google and Bing shows that Bing still cannot understand these kinds of queries. This is a fact. Not opinion.

This is the latest and up to date state of search. If you're still obsessing about Rank Brain you really need to set that aside and catch up to what Google is doing right now.

Neural Matching is very likely among the biggest changes to the algorithm alongside Penguin, Panda, and Hummingbird.

[aristotle]

I just did some searching in another sector ( information about a particular social issue), and the quality of google's search results in that sector has also gotten worse -- and for the same reason -- In one case four of the first seven organic results weren't very relevant to my search term.

From what I've seen this problem could be affecting virtually all of google's results, although some sectors could be more vulnerable to the negative effects than others.

But as has been pointed out by other posters, it may not have a big enough effect on most searches for the average person to notice that the results have gotten worse. At least not yet.

[JS_Harris]

I think it's a bit different than we suspect.

To me I see Google's "odd" behavior at times as guessing what I mean and putting me on the wrong part of the "tree" of answers.

Google tries to show you the most relevent stuff but if you look at questions and you reload them over and over you can see that they become much less accurate from your original query over time, and more accurate to variations. When they put you in the wrong branch it feels like synonyms but it's really just not having understoof the query, perhaps because it had many meanings.

[martinibuster]

>>>Google tries to show you the most relevent stuff...

The word relevance can be inaccurate because it means a lot of different things. It can mean matching keywords to content.

What Google is doing now it showing the most popular answers. Google says they're showing the most useful, but only within the context of what is the most popular definition of usefulness.

;)

Roger Montti

[broccoli]

I'm finding it rather annoying when I do a search and a third of the results show up that don't include one of the actual words I typed into the box, and after scrolling for a bit I have to click "must include" to get what I was really looking for. I seem to remember Google did this a few years ago and it didn't work out then either.

I'm rather concerned about the use of machine learning for this kind of thing. Companies have tried to use machine learning to remove the possibility of human discrimination (regarding race, gender, etc) from fields such as selecting interview candidates from CV submissions. Studies of the results have found that the software develops much stronger biases than humans, because it sees correlations but can't see past them or correct for them and make exceptions, as humans can. Machine learning ignores exceptional candidates from outside the box and discriminates towards the standard applicant, e.g. towards white male engineers or black female nurses, and actually has to be programmed to ignore those factors so individuals can be assessed on their own worth, not towards a standard measurement of the herd. The problem is there are probably infinite factors that individuals should not be assessed on that could correlate in some spurious way.

If Google is doing the same with neural matching, it's going to filter out or deprioritise all of the exceptional and interesting websites before they even get seen, because they don't check the standard boxes. That's particularly concerning where the overall standard of a niche is below par. Are you going to filter out the best websites to answer the query because their content is skewed somewhat from the standard answer? What if those websites have new or better information than the rest of the herd? Ambiguous queries require a greater diversity of results, not a best guess based on what the average person thinks they want because of what they think they know. That's how you end up in a bubble.

(Of course I say this as someone who has an outside the box website, who experienced a 30% drop in traffic over the early part of this year...)

[martinibuster]

it's going to filter out or deprioritise all of the exceptional and interesting websites before they even get seen, because they don't check the standard boxes

Actually, Google's algo does the opposite. It includes all the exceptional and interesting websites that don't check the standard SEO boxes.

So, rather than set aside a great site that answers a question in a meaningful way (because it doesn't have the exact match keyword in the title tag), the algo will include those sites for consideration in the SERPs.

WebmasterWorld has come a long way
I'm glad to see the community finally admitting that synonyms are in the SERPs. I tried having this conversation with this community a few years ago and encountered stubborn resistance from some who refused to let go of their Old School SEO beliefs. Even after they privately gave me keywords and I showed them with their own keywords that keywords weren't necessary to rank, they refused to believe what they could see with their own eyes in the SERPs.

But there still needs to be some catching up. This whole synonym thing is the conversation this community should have been having a few years ago. Google is years beyond synonyms and rank brain.

I'll repeat what I previously said, what Google is doing now with neural matching, is in my opinion, one of the biggest changes to the algorithm since Penguin, Panda and Hummingbird. This is important to understand because it is part of this conversation about Synonyms.

;)
Roger Montti

[broccoli]

Actually, Google's algo does the opposite. It includes all the exceptional and interesting websites that don't check the standard SEO boxes.

I'd love to believe you, but I just don't see that. When I see the same bad actor who checked all the SEO boxes but gives users unsatisfactory answers sat right at the top of my niche, getting rewarded all the way down for all of its keywords, and frequently appearing for long tail keywords it isn't even relevant for, what I see is a one size fits all approach. The same people are even using keyword manipulation to rank a second off-topic site above me. My "niche within a niche" site has been hammered by Google this year, and I've been forced to add content that isn't relevant to what I actually do in order to claw back my long tail searches from the "authority" sites that sit at the top of the niche.

[broccoli]

By the way, thank you @martinibuster for the article you posted last month on neural matching. I just re-read it and what went wrong for me really does make sense in this context, if they aren't counting backlinks in these types of queries. I have the second biggest backlink profile in my niche yet it feels like my site has been vanished by Google.

It explains why a "poor quality, bad actor but checked the SEO boxes" site is dominating my niche without a significant backlink profile, since the short tail keyword is ambiguous. I still rank very highly for the short tail in every other search engine. I wonder whether Google are also ignoring Chrome user experience data for these queries, or even whether Google is measuring my user experience correctly, since my pages contain ajax.

It would explain why I lost a lot of traffic earlier this year yet I just couldn't find where on earth my rankings had dropped for my keywords. I had been (I think) on the second page or bottom of the first page for the short tail keyword, and that dropped through the floor. But the majority of traffic seemed to disappear from the very long tail/unique queries that are impossible to track. I also used to get a lot of traffic from queries like "funny [my keyword]" and "cool [my keyword]", which was fair, because that's what my site is known for. I got replaced by the bad actor in all of those kinds of queries, even though they have possibly the least funny site in the niche. I thought I'd been given an algorithmic penalty for something and blamed myself, but maybe that wasn't the case.

Neural matching just doesn't work well for my niche, because I have a kind of tool/game/quiz site. Most sites have virtually no written content on the page (I'm actually an exception in that I try to get at least 400 genuinely informative words on each page - I'm now concerned that's counting against me and I look like the odd one out and therefore less "relevant"). The quality of the page doesn't relate to the written content however, it relates to what is produced by the tool/game/quiz, which isn't something Google can measure on the page. So Google is basically rewarding sites that have put their energy into technical SEO, and ignoring user's evaluations of those sites.

This makes me feel pretty depressed, because it means all the work I've done over the years to build a reputation, focus on user experience, and get people to enjoy my site enough to give me natural backlinks has gone to waste. It also means my traffic share is going to be very vulnerable to attack in the future even if I can get it back. If that's the direction Google is going in, then there's no point me putting any more work in to making my site better for users, which is where I've been focussing most of my efforts the last few months.

However, if all that Google wants is keywords and synonyms, then I guess there is something technical I can focus on to hopefully make some kind of short term difference.

I wish there was a way I could get this kind of feedback through to the Google developers. I've tried asking for help and posting on the Google product forums, but I never get a sensible response, just the brush off with cookie cutter comments about EAT (even though my EAT is better than any other site in the niche, and my site actually got a slight traffic increase after 1st Aug).