Do any of you guys apply the principle of semantics / related words on your pages in a bid to increase rankings for your main keyword? I have heard (I know we hear a lot but can never tell for sure) that G's algorithms get better and better with the whole language/semantics and neural networks deal. Stemming and plural vs singular is just part of it but do you reckon G can evaluate what your page is about by looking for related words i.e. semantics? It makes sense for determining relevancy but from a technology point of view it seems pretty hard to me to program. Every word then would need its own database of related words and perhaps even ranked in/weighed on their relevancy to that one particular word. That would create a virtually infinite web of related tables wouldn't it? It would work against spamming and bombing though. If a site would be optimized for the phrase 'motor bike' and it also mentiones tons of related words like race track, helmet, racing, championship, tyre, brake, crash, lap, crowd, engine, horsepower etc etc. it would indicate a relevant site. It also works against keyword stuffing overkill. To what extent you reckon this is already being applied? How smart are the bots when it comes to language?
Google is alrady using stemming, and other search engines use clustering which is eally an application of semantics. IMO (and this is really me just thinking on my feet) Google would probably build its lexicon fom a combination of a backend such as rogiets or the like. and compare its search data from there. A working example would be that I am searching for say bike helmets. I get a load of push bike helmets, so I right away type in Mortorcycle helmets. Google would know I am repeating a search, and could compare my searches to its back end control. it would indeed find that bike & motorcycle 'can' be stemmed words. I would think that Google would build their own lexicon in this way, rather than using some giant book of words, the bulk of which would never be used. The big draw back in all of this though, would be that Google would be giving bad results, and it would take a while for the system to build its lexicon, and correct it based on other words within the search. Google apparently did toy with stemming during Florida, although so much was happening that it was hard to tell black from white.
to some extent they can already do that on google. a search for SEO will give one set of results. a search for ~SEO will include releated words.
Interesting! I wasn't aware of that search tool. It seems pretty clever on abbreviations and nouns vs verbs. This must be the same as how they 'broad match' adword keywords. I normally like to stay in charge myself though... I guess, with the millions of searches they get, building its own lexicon is less hassle than I initially imagined.
I read a research article over at goRank.com ( http://www.gorank.com/research/google_ontology_analysis.php ) they call it Ontology. Basically their findings showed that pages, in Google's SERPs, ranked generally higher if there are more related terms/keywords on the page to the keyphrase that was searched then you will generally rank higher. Of course this is hard to tell for sure, but it makes sense, like you said. They did make a tool to help with this type of thing. http://www.gorank.com/seotools/ontology/ And doing a short little test it seems to be true. First I searched just the word media in Google. The first result is indimedia.org with the word media bolded in the right spots. Then I searched ~media and the first result is real.com with the words audio and video bolded out. the word media isn't even there let alone bolded. Then when you type in the word media into that Ontology tool I mentioned above you see that video and audio are the first words that relate to the word media besides media itself. This simple test and the research GoRank did seems to show that yes Google does do this in one form or another. how they do it is an entirely different questions though. Of course I am somewhat repeating what vagrant showed us in the post above mine. I know when I write a page from now on I will see what words are related to the keyword I am trying to optimize for just for that extra boost. - Sean
I find that doing a search for ~keyword ~phrase is a good way of finding new keywords to target and often ones that i would never have thought about. Never spotted it mentioned on an SEO forum before tho, so i assumed it was just something everyone knew about.... but could also just be that i have never read the posts that talk about using a ~ in front of a word on google.
If you check with that Ontology tool it says that the only words related to "keyword" are keyword and keywords. And if you search "phrase" there are no related words. So apparently not all words have related keywords to go with them. Maybe the database isn't completely created? Maybe the database only does the top 2000 words searched? Maybe it's Maybellineâ„¢? I was branded as a child by multi-million dollar companies what can I say?
Great thread, and quite relevant. Perhaps an hourly kw density analysis of this thread will yield some primaries, then relateds might narrow the field a little, such that adsense realizes a 'G-Tech Performance Meter' & 'Supercharge any Car' are not really consistent with the theme.. Of course there's more overhead than an 'archive' but still, there be bikkies here !
I just tried some british spellings vs american ways of spelling and G seems to know which are the ones that mean the same. Try ~neighbour, it will come up with neighbourhood, neighbor and neighborhood results. To a lesser extent when you try ~neighbor, probably because on the net that is the most common way of spelling it. BUT... Trying ~optimisation it doesn't show optimization. Not very consistent it seems yet there might be a difference in handling vowels & diphthongs and consonants for some reason.
I think this has more to do with Google's Stemming abilities than Semantics... It will get better, I'm sure - Early days yet.
I agree. Google only seems to get better with age. These handy little search queries are the proof. My favorite is how Google can convert units of measurement easier and faster than I could do my own math, if I actually remembered formulas. Things like " 357 miles to meters" and less than a second Google spits back info from a data cluster further away ... 357 miles = 574 535.808 meters Thank you Google, and keep up the good work.
I am thinking of setting up a little test to see whether Google does rank page A for key phrase X higher than page B for the same key phrase X if indeed page A contains a lot of words related to key phrase X but page B not so many. I will use my own imagination when it comes to the related words; using the ~ search or AdWords suggestion might be too obvious. (or maybe 3 pages of which one loaded with words stemmed off phrase X, or words from phrase X, as given by google through the ~ search or AdWords suggestions) I will make sure the actual key phrase X has the same density on both pages and all on and off page SEO are the exact same too. The different file names will have the most insignificant difference I can think of. I found a not-so-ridiculous key phrase which now only shows 8 results world wide so it shouldn't be hard to get in the SERPs soon. Before I get my behind into gear... Does anyone think this is the way to do it and do you think it will be a useful test? Anything I forgot?
I posted this message http://forums.digitalpoint.com/showpost.php?p=2069&postcount=30 in a thread in early April that tells you what I think and about Google's use of semantics and like tools. It also contains a pretty good directory to papers on all these subjects. You might want to take a look.
I found this really helpful in understanding semantics. Anyone who is focusing on it or considering should probablytake a look at it: Patterns in Unstructured Data With the ~search I do it this way: ~apples -apples then it mentions 'fruit' so I search ~apples -apples -fruit then it mentions 'oranges' so... ~apples -apples -fruit -oranges then 'cherries' and so on. Get your list of related words then repeat on each.... Chop the fruit, pour on custard, click on digitalpoint and check out the lattest banter on the google update
Yeah, this is a great article... Read it a couple of times now... Still pretty hard to get your head around though... Not sure how much weight is being put onto semantics at the moment though... I have sites running that are doing ok in the SERP's but are definetley not "semantically correct"...
I have set up the test a few days ago and have some results now. I will give you all links etc. later after I had a closer look myself in the evening. Three pages have been made with each the exact same number of words and characters. Each of them contain the key phrase 'kw1 kw2 kw3' exactly 6 times in the paragraph text. On and off page SEO is the exact same. The only difference is the content to see if G understands what the pages are on about, and if so, does it affect ranking? Page X has pretty much random content, not related to the targeted key phrase at all. Page Y has lots of words related to the use of colema boards, as per my own imagination. Page Z is stuffed with words related to the keyphrase subject as suggested by G's very own Keyword Suggestion tool. If it does understand semantics and knows which words are related to which other words, then these suggestions are likely to come from the lexicon they have already built. Now the shocking first results: 1: Page X (with B.S. text!) 2: Page Z (with related words) 3: Page Y (with G's suggestions) The page I made to get these three spidered does have Page X as the first link so maybe that might skew the results??? I'll let you see the pages later today after I had some time to check it al out...
More or less takes originality to the extreme doesn't it ! I'll be interested to examine the relationships