Recently I completed a major project – a standalone Macintosh application called “Myallo” (a name based on a Greek word meaning “brain”) that used neural network techniques in order to do web searches and rank the results much better than any online search engines could.
If a traditional search engine pulls up a million hits on “gardening”, how does it determine which ones to put at the top of the list? It only has a few things it can do; estimate general popularity by counting how many other sites link to it, see how many matches to the word ‘gardening’ are on the page, check the locations of those matches in the on the page and so on.
But if I gave a program more info on what I’m interested in: gardening yes, but I’m more interested in vegetable gardening than flower gardening. Now it can rank these hits according to my own actual interests!
So in the stand-alone program, you first create an Interest Profile based on a hierarchy of topics. The application can then go to a search engine to pick up hits, then scan the contents of some of those pages to see how they match up with all of the user’s interests.
This works well, but the application was restricted to Mac users, and the user had to set up a profile of all his interests manually. So I determined that making this a website could be good. Any computer user could utilize it, it could scan a large number of articles from around the Web for all site members simultaneously, and by using a large set of topics common to all site members, it could be made so a user wouldn’t have to set up a profile, rather they could just give some feedback on the results, and the profile can be gleaned from that.
The new site, “Myallo Online” at http://www.myallo.com doesn’t have the user doing any searching at all, and they don’t make up an Interest Profile either. The web site does the searching and interest-tracking, all you do is rate how interesting some of the articles are, and since it pre-scanned articles to see what topics they are relevant toward, it can compute how interested you are in those topics, and makes up your Interest Profile automatically. You can still see and adjust the profile directly, but it’s not necessary. If you rate a bunch of articles about vegetable gardening very high, but those on flower gardening only moderately high, it figures out you are quite interested in gardening, but more in vegetable than flower gardening.
Instead of asking for search terms and running them through search engines, the site just goes out and picks up lots of RSS feeds from around the Web and scans them to see how interested you might be in them. Then, when you come to the main page, it shows you the most interesting stuff first.
Basically, you do no searching and no setting up, yet you get a sort of constantly updated newspaper that automatically gets smarter about what kinds of things should be on your customized front page. As you tweak things by setting a slider next to the article to say this story was more (or less) interesting than Myallo Online predicted, it literally learns and improves.
So i think it’s a pretty cool concept, it’s up as a pretty nice looking website, all it needs is a little traffic now. But that’s the real trick, isn’t it…