Google's Semantic Processing Is Getting Very Savvy

[tedster]

< Note: In another thread [webmasterworld.com] about title tag rewriting,
Sgt_Kickaxe and I touched on a related topic that is worth a dedicated thread. >

For a long we've seen Google move away from exact text match and into more sophisticated semantic processing. In recent weeks they seem to have made a big leap. Here's what I'm seeing:

1. URLs ranking really well with only synonyms of the terms that are in the query phrase

2. Display titles getting rewritten to use the related keywords instead of the original

3. Suggestions sometimes use semantic equivalents, rather than just based on the text that is typed in the search box

I'm not talking just about one word synonyms, either. I'm seeing equivalent PHRASES being plugged in, and in a way that makes plenty of sense. This kind of shift can have major implications:

A. Traffic for keywords you never thought about

B. Search volumes for keywords might change a lot over what the Adwords tool reports. This would be because sometimes semantically related Suggestions and even Instant results might hijack the original search idea

C. Semantic intelligence like this could disambiguate the searcher's intentions, resulting in better targeted traffic even though the volume goes down.

It's easy to focus on what Google is doing wrong in any moment, but at times I am quite blown away by what they can do unexpectedly well - and this was such a moment. Has anyone else noticed this kind of change?

[Planet13]

Has anyone else noticed this kind of change?

Yes. But it almost seemed to be outsmarting itself.

I won't use the exact keywords here, because it mentions a product, but basically I was looking to purchase a service to digitally copyright my images, but I wanted to see if that service was easily cracked because I didn't want to spend money on something that wasn't going to work. So what I searched for (with an alteration to the keywords so as not to mention the product by name in this forum) was along the lines of:

"defeat digital marking"

Now, almost all of the top 10 results were for the company that sells this service. (Not indented - just multiple regular results).

Then there was a listing for a DJ with the name of marc, and the word "beats" (as in musical beats) was bold in the snippet, despite not being in my query.

So in this example, we saw the NAME Marc being a synonym for the noun mark (as in watermark or mark up), but more interesting to me was the word defeat was transposed to "beats" (a reasonable jump) but that it was referencing musical beats as opposed to the related meaning of "to conquer"

So either the alg thinks that ALL instances of the word beats are synonyms of defeat, (we can probably test that out easily), or there must be some trigger that expands the scope of the matching synonyms when there is a dearth of pages that mention ALL the words on the same page.

~~~~

Also, I wonder if anchor text from backlinks influences how well a site will rank for a synonym.

Supppose we have a site that is about widgets, and DOESN'T mention the word foo bar on there.

But we build a huge number of inbound links that use foo bar as the anchor text, even though it is not a common synonym of widgets.

Could we get that widget page to rank for foo bar based solely on the anchor text of inbound links? Could we get google to bold the word widgets in tne snippet when a searcher looks for foo bar?

(In reality, the page will probably be penalized - but I give this as an example to try and get ideas on how google LEARNS of synonyms.)

[Sgt_Kickaxe]

Simple test, search for something using 2 or 3 or 4 keywords like you normally do and count how many results have the EXACT words you searched for ONLY. I'm really not seeing many even for extremely popular subjects like tech gadgets etc. I remember always seeing at least one and usually 2 or 3. Top spot was almost always dominated by a site with exact match domain even...

It takes time for serps to finish propagating after a change is made in the algo, we might just be noticing this lately because more pages have finished propagating the change and it's more noticeable. Google announced a change that "will affect 12% of searches" and "will affect longtail" etc some months ago now, this might be it?

edit: I wouldn't go on a mad title changing spree, instead I'd avoid the temptation of using exact searched for terms exclusively, ie: add other descriptive words that match the page, and build a good mix of both.

[shazam]

edit: I wouldn't go on a mad title changing spree, instead I'd avoid the temptation of using exact searched for terms exclusively, ie: add other descriptive words that match the page, and build a good mix of both.

I agree.

With all the focus falling on over-optimization, synonyms in the title may be more effective. Check the top results for your terms and view through LSI colored lenses to get a glimpse of what they're up to.

Personally I'd prefer a search engine to return exactly what I queried for, not what it thinks I want.

[jmccormac]

So Google is doing the equivalent of what the scrapers are doing with spinning text? How ironic. :)

Regards...jmcc

[robzilla]

ie: add other descriptive words that match the page, and build a good mix of both.

i.e. write naturally ;-)

Personally I'd prefer a search engine to return exactly what I queried for

Not me. I can't count the number of times Google saved me by including results with keywords I would never have thought of myself, but were nevertheless exactly what I was looking for -- that's impressive engineering.

[g1smd]

If Google spins the text and the scrapers spin the text, does that cancel everything out and demote the scrapers? I surely hope so.

[Lapizuli]

I've noticed it, too. I look at semantic indexing as basically a fancy name for artificial intelligence.

Google's headed there, but is kind of at the new toddler stage - it's discovered it can walk, but it really can't get very far, so it holds up its followers while it toddles clumsily around.

To use another simile (for fans of the late great Douglas Adams), of late, Google's latest targeting has been like Arthur Dent's attempt to make tea - the computer will give its all to make the perfect cup of tea for one man at the cost of many lives.

Basically, Google results have shown diminished AI capacity in our household over the last few months - it's noticeably better at pinning mainstream interpretations on searches, but worse at long tail searches. Right now, I can only find sites in Google that I'm specifically looking for - discovery searches are giving me big brands only, or spam in niches that don't have big brands attached to them. And quite often I can't find things I found before. A few years ago we gave up bookmarking sites we visited. We're once again a fully bookmarking household, with half its members converted to Yahoo!.

[supercyberbob]

Lapizuli

Very interesting. I was hesitating at mentioning it, but I've been having major problems finding what I'm looking for on Google these days as well. Good to know it isn't just me.

We should probably take a good hard look at whether search quality has improved since Panda.

Can anybody honestly say there has been a significant improvement in search quality? Feel free to chime in.

[brinked]

tedster,

I agree. They have gotten much better at figuring out what the user is trying to find. However, there are still moments where I am left smacking myself in the head that google can find some webpages relevant for my query.

I have high expectations for google, sometimes maybe too high and this is something I always envisioned google excelling at. It has much improved but it still has much room for improvement.

This is something that can never truly be perfected but it will go a long way the more google can improve on this.

[tedster]

It depends what you mean by "recently". Very recently, as in the past 2-3 weeks, I'm having a significantly better time finding things with Google. Earlier this spring there did seem to be more of a problem.

Google's Information Retrieval science seems to be in a very assertive phase. They're taking on monster tasks that no search engine ever tried before. Sometimes that has a negative impact for my little corner of the web, but over a broad sweep, I think it's a powerful net gain.

I like the toddler analogy. And let's face it, the average Google user didn't really understand purist search results, so more advanced methods are certainly understandable. Guess that means putting up with temporary misfires along the way.

At any rate, for now I wanted to acknowledge this improvement. I see it in the server logs for sites I work with, and I see it in the personal and professional searches I do.

[walkman]

I'm sure from an engineering perspective this is interesting and challenging. But personally I don't like this or personalized search. I understand that offering 'suggestions' is great but I don't want anyone to guess what I'm thinking at all. Maybe showing me what I searched for first and then show whatever they think I wanted to search for. This applies to FB, Twitter and +1 suggestions from my 'friends' as well.

To me doing a second search, scrolling down or adding an extra keyword isn't the end of the world.

[outland88]

It seems to benefit the brands very well especially the major brands. I really don't believe it's meant to enhance the fortunes of small business. It's a sly method of favoring big business.

[kellyman]

It depends what you mean by "recently". Very recently, as in the past 2-3 weeks, I'm having a significantly better time finding things with Google. Earlier this spring there did seem to be more of a problem.

Well that's a very good point, there's another thread here in regard to panda 2.2, my traffic is down this last 10 days or so, with no real ranking changes, but this makes perfect sense now

Traffic down, Visitors down but sales stable, maybe Google is getting so smart in returning results, if this is the case it would result in less long tail slightly relevant content and give you slightly less results but better quality

I think the days of people ranking 1 page for many keywords have gone and Google is making people spend time on their sites, i have had some of my article pages returned when Google used to return the home page and makes sense to do that

I know its been doing this for a while but something has changed this last few weeks with no real idea of exactly what and it looks like the emphasis is going down the quality route in a big way, i know its been trying to do this for ages but this new roll out could signify and very big push in that direction

[tedster]

I'm toying with a similar idea, kellyman. Changes in semantic processing may be an unacknowledged "ghost in the machine" for recent ranking changes.

[netmeg]

My (small business) clients sell products that can be legitimately referred to several different ways (and in many cases the estimated searches are about equal) so yea, we've noticed it, and we're really happy about it. Because often the only real way to get some of the alternatives onto the page or into the page titles is to keyword stuff like crazy.