h a l f b a k e r ySuperficial Intelligence
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Lots of sales websites use a "People who bought X also bought Y" recommendations algorithm. But what that doesn't factor in is change over time.
Tastes mature. For example, person Z might start off liking Evanescence, expand from there to liking the goth metal sound of Within Temptation, from there
to symphonic metal like Opeth and maybe even drift into some Death metal like Archenemy.
A traditional recommendation system would recommend Archenemy to people who like Evanescence - even though they're 'not ready' for it. Conversely, Archenemy fans will be recommended Evanescence - even though person Z may no longer even like Evanescence.
Factoring in how far away Evanescence is from Archenemy timewise avoids this problem. I'd also weight in favour of forward-moving time. (In the example above, Within Temptation would be strongly recommended to Evanescence fans, but Evanescence would be recommended less strongly to Within Temptation fans.
The beauty of this algorithm is that most sales sites probably already have the data required to put it into effect. Amazon.com has been around for 14 years, and it's recorded the date on which every sale was made. That's a database of 14 years of evolving tastes.
Netflix Prize
http://www.netflixprize.com/
Still going, but I expect them to finish some time soon - offering one million USD for 10% improvement, have 9.15% [jutta, Aug 17 2008]
Please log in.
If you're not logged in,
you can see what this page
looks like, but you will
not be able to add anything.
Annotation:
|
| |
good, practical idea. How is this halfbaked? :p |
|
| |
I've long been of the opinion that each person's taste in music (and perhaps other art) is a unique construct, a product of the path they have taken through all the music they've heard. Perhaps this would render other people's paths less useful in determining your future path, perhaps not. |
|
| |
Still, I like recommendation systems and any way to make them better is worth trying. |
|
| |
You know, NetFlix offered a prize for folks who could improve their recommendation system. They supplied anonymized recommendation data to test algorithms on. Maybe this could be tried on that. |
|
| |
I like this idea: "Analysis of your buying patterns indicates that you're just about able to give the impression of being a bit avant guarde - you'll probably buy the new Sigur Rós album next - but in ten years time all those CDs will be gathering dust and you'll be into 'Dad Rock', so why bother? Why not just give up now?" |
|
| |
// Consequently, you might also like "Paris" by Paris Hilton. // |
|
| |
Obligatory "We'll always have Paris" quote. |
|
| |
// Why not just give up now? // |
|
| |
Because we will be able to sell the CDs on eBay. |
|
| |
This whole idea sounds like a sort of "six degrees of separation" algorithm (unless anyone would like to become One with the Borg ? We have a special offer this month .......) |
|
| |
Would be interesting if you let your kids use your account. "People who bought this album also suffered a nervous breakdown and bought all this crap." |
|
| |
What does this mean? //...how far away Evanescence is from Archenemy timewise...// |
|
| |
I'm not familiar with either band. Are you talking about, oh, Evanescence is from the 80s and Archenemy is from the 90s? (I'm just using those dates as examples; I have no idea when these bands actually existed). |
|
| |
I love Paris in the Spring time. |
|
| |
One of the challenges is to recognize the trend without creating too many false positives. In the context of music/movie purchases, a number of items could be bought as gifts and therefore not actually part of a person's tastes. |
|
| |
I think that a good example of this technology is currently used by the guys over at Last.fm. Their underlying application (which they built) for the recommendations is called 'audioscrobbler' and it is pretty good at recognising genres and suggesting stuff for you to listen to. |
|
| |
Furthermore, in the context of overall context-aware applications, this kind of AI (which I am interested in but not an expert) is a very big research topic right now. Not least because many currently important companies have business models based on targeted advertising. "You bought X ... so here is XXX!!!" |
|
| |
Did you mean: Al Gore-ism? |
|
| |
[+]. But if I ever get recommended Evanescence as a result of this idea, I'm coming for you. |
|
| |