[encoderer]

The core of these recommendation algorithms is often a flavor of nearest neighbor analysis. In this case, the computer is using attributes to find people who are similar to you, and then suggest to you things that they liked.

All the "robots" do is find people who it thinks you would like to emulate. That sounds like the same thing the fashion industry has been doing for decades.

[jfarmer]

Robots don't merchandise, that's the problem. For many people decisions about what to wear is very intimate, but algorithms are bloodless.

The who, where, and how matter much more in fashion. Algorithms miss that.

[encoderer]

I'm not really opposed to what you're saying, but it does seem to me like you dismiss the big human factor in these algorithms. It's not like it's just a system saying "You bought a red dress last year, here is a list of other red dresses you may like" -- Because the quirky frock she bought for work last year doesn't mean she wants to see cocktail dresses that just so happen to be the same color, or from the same designer, or in the same price range.

Instead, it's correlating prototypical woman A to prototypical woman B based on available demographic data and past purchase history, and suggesting to "A" some of the things that "B" purchased. The emotion and intimacy there was the human act of "B" purchasing clothing that she connected with.

[div]

I think a strong algorithm may surprise you.

I remember reading about the Netflix contest and how one of the algorithms categorizes movies into it's own statistically relevant categories*.

Once a comparative algorithm is used and the dataset gets large enough, I have no doubt the recommendations will start to get real good real fast.

http://devlicio.us/blogs/billy_mccafferty/archive/2007/01/02...

[jfarmer]

What I'm trying to say is this: it's not about a "match" between the product and the potential consumer.

The who, what, and how of the recommendation matter almost as much as the product itself. Just the fact that I know this recommendation came from a human counts for something, especially if it's a human I trust or respect when it comes to fashion (not necessarily a friend).

e.g., a celebrity wearing a shirt and having it sell out the next day.

Movies are different because their consumption isn't inherently conspicuous. I do it alone and talk about it with friends if I choose, but everyone sees the clothes I wear no matter what.

For example, I can choose to hide the fact that I love Katy Perry, but I can't hide the fact that LVMH made my handbag.

How you dress is a performance, and so the decision to wear something is filtered more rigorously through a social dimension than watching a movie or listening to a song is.

[encoderer]

In the context of my most previous reply above -- how would you feel about reccomendation algorithms that pull more visibly from your own social graph -- Friends, yes, but also people you follow on Twitter and Like on facebook.

That seems almost a perfect marriage of my comments on the innate human quality behind a nearest-neighbor algorithm your comments here.