They are collected from the "People who visit this page also visit:"
I Would guess they call the last site visited which matches some keywords, and then add it.
I think in most cases it's not very easy to change, unless the site isn't very popular.
Well, I have a pr6 site about fish, that has links to a car page.
There will always be glitches, there is some huge glitches when it comes to .co.uk domains, but i think it's just to wait, or visit a lot of similar sites, and then return to your site.
If you're the only person (or one of a few) with an alexa toolbar who visits a page then you can get some really strange results on the suggested links. On an internal (pword and ip protected) page we once had "other people who visit this site also went to: fark.com webemasterworld.com theregister.co.uk"...
Turns out that I was the only person to ever use this page and it was just picking up the links I came in from...
hehe I stopped using it because it was suggesting the people go to my competitors pages! :)
Guess where I had been spending my time....
When I was using it it looked like it found my ODP listing category and displayed sites from that cat. I had assumed it was all drawn from ODP.
But, now that you mention it, it might be that somebody with the toolbar had been working their way down the sites in that category.
Looks like you are too popular to influence those links, Brett
(if the above theory of similar user visits is true - on the other hand you could ask all members to put on an experiment..)
What I find strange is that whilst more than 80% of my visitors are not Dutch, 90% of all suggested links are Dutch.
In a way, once the links are suggested, they would tend to reinforce their placement, purely by the fact that they show up and ask to be clicked on.
My links were also showing my competitors pages! So I stopped going to those sites using my computer which has the Alexa bar!
I think it's not dissimilar to the Amazon "People who bought this also bought..." For high volume titles, it works pretty well. For low volume titles, the results can be irrelevant, e.g., an Italian cookbook suggested to a purchaser of an XML guide. Didn't someone do a Googlebomb-style attack on this a few months ago and get Amazon to suggest some kind of perverted book to buyers of a book by a TV evangelist?
I share Brett's feeling about the seeming randomness of the Alexa site suggestions. Sometimes they seem quite logical, other times kind of mystifying, e.g., low volume sites instead of the logical, higher volume ones that you know must have common visits.
I suppose Brett could conduct an experiment - post a link to a desirable partner site here, then suggest we turn on our toolbars and click it. An experiment would no doubt be easier on a lower volume site, though, where a couple of hundred clicks on a related site would overwhelm whatever junk was in the database.
i think it is really tricky to pull this kind of functionality off and always get it 100% right.
a couple of products / modules i have developed based on search behaviour get *really* erratic when they are dealing with low volumes of data, and i guess it is the same for any stats processing, which is effectively what this is doing.
an off-beam result doesn't matter when you are dealing with a large dataset, but can completely screw the output of a small dataset.
Lots of excellent guesses! Alexa actually reveals the methods via FAQs 'n' things:
1) People who visit this page also visit. This is the same technology used by Amazon.com, originally written by Alexa. Because of the massive volume of this data it only gets updated every two months or so. Generally you wouldn't have to worry about your own "paths" generating Related Links because more than one unique user needs to make the path before it is deemed relevant.
2) Suggestions. Users can suggest links via a form on the bottom of the detail pages. Sometimes these get spammed, but Alexa Editors review all the suggestions and try to clean out all the junk.
3) Links found on the web (including DMOZ, Yahoo, and the rest of the web.) Basically, if two links appear side-by-side on the web, they are deemed to have a relationship. Relevancy goes up if the relationship is confirmed on multiple sites. Sometimes these can be weird relationships(have you seen some of the categories in DMOZ and Yahoo?)
4) Editors. Sometimes people just need to get involved and make links happen.
Lastly, if there are numerous links, they are sorted by increasing the relevance of links that are found by more than one of the methods above.
If you need to change links that appear in relation to your page you can use the suggestion form. But there's no guarantee they will ever show up. If there's some link that is particularly irrelevant or unflattering you can contact email@example.com and they can take some action.
>gets updated every two months or so.
Ok hmm. I can see that is alot of data to crunch. Wish that was faster.
Thanks for the info.
I played around with a music/film site called Firefly 3 or 4 years ago, or maybe even 5. You rated how much you liked diff movies and CDs by cliking on them, and soon, everytime you logged in you got the "you may also like [list]". it did take a long time for the program to really "work you out". At the time firefly was really the only people doing this sort of thing. Im pretty sure I heard that Amazon bought the technology a few years back, and a bit later it started to be evident in Amazon. I guess with websites it takes even longer to get reliable info, the unit of anlaysis not being the individual (tho Alexa does this also with "the web you made" I guess in a rudimentary or testing fashion), but the site.
Ok, that old "suggested" links page where you can suggest links is still there. I thought that was gone is why I asked in the first place.