« endless loop | main | go sign up »
Love Amazon? Check out Rashmi's post to better understand your loved one...
'... it is incorrect to think that Recommender Systems cannot have an agenda, or less of an agenda than categorization. Recommender Systems are explicitly designed to encourage people to buy. ... Apart from other things, they also classify YOU. And they classify you without any knowledge or choice on your part. "
I work for a large business that has a history of being 'evil' on the odd occasion-- it's depressing but it is a good company in general and its failings should be called out-- as should other companies. If I were going whole heartedly after big companies that do bad things, I would certainly be in a glass house.
As for Rashmi's post it was well written and a compelling argument and worth blogging. Do I agree with all her points? Mixed feelings. I love Amazon and spend a lot of time on it. I hate Amazon and spend a lot of time on it. The gold box is a joke. The unedited user reviewes are a godsend. But at no point will I ever turn a blind eye to a comapny's failings because of their good, or their good because of their failings-- unless their failings reach a level of horror of course. But if we hope to excel in our craft, we must look at it all-- the good and the bad, and then make up our own minds.
Is Amazon evil? No, and I never said so-- and genuinely don't think so. I sure hope they don't turn out to be-- where will I buy my french music or esoteric books?
BTW "She also states that metadata search & browse is inherently superior as it empowers users with choice and freedom of anonymity. " is inaccurate... reread the post. I just did. I found it remarkably even handed. She merely points out that recommender systems also have bias and agendas, as do other classification systems. Moreover, if you read the entire thread, i think she is merely warning us not to take things for granted... I think this is wise advice.
Finally, this blog is run by a human being, who will on occasion be human, sorry. I am not a journalist, and I am far from unbiased. Less biased than say, the san francisco examiner, and more than the new york times. But still, full of opinions (though lately too tired from long hours for agendas... maybe later)
>>>
Better Recommender Systems , which truly help
users find new items, and expand their tastes, find much less of a market. ...The story is the same, many
Recommender System companies are finding it hard to sell their best
products that really helped users explore their tastes. The Recommender
Systems that sell, are ones that help companies make the sale, and make it
fast.
>>>
I have some trouble believing this. I have some trouble believing there is much of a gap between helping a user discover their tastes and helping make a sale. Rashmi doesn't go into detail on this point, but I would like to hear more. How does the gap open up? Since I'm having trouble imagining it, could an example be given? It seems like any time you can introduce the customer to something they will like, you are increasing sales.
Does Rashmi have a weblog?
Lawrence, this is the paper that I was referring to on the post, which describes the dissociation between the Recommender System that users loved (MediaUnbound) and the one they bought from (Amazon). The dissociation is primarily based on familiarity. Users like learning new aspects of their tastes, but they do not want to spend money on new and experimental recommendations. For example, following up on my discovery Latin Jazz (through a recommender system). I did not immediately start buying such CD's. No, it was a little while before I spent money buying cd's of Latin Jazz. Over the next few months, I made sure that I did indeed like Latin Jazz. In contrast, Amazon might recommend a second cd by Sarah McLachlan, when I already have one in my shopping cart. I am highly likely to buy this second cd. This does not tell me anything new about my taste, but it does help make the sale. I believe there is an inherent contradiction between being a good recommender system and a good ecommerce system (at least in the short term). Amazon probably knows and understands this and has fine tuned its algorithms to stay very close to the user's initial choice, rather than be experimental.
This was obvious time and again in our studies. Users were annoyed by Amazon recommendations (Amazon often repeats the item that you just searched for among its top recommendations!). But time after again, they said they would buy Amazon recommdations.
Good recommender systems (if you define good as helping you learn more about your tastes) take 5-10 (even 15) minutes. In this age of instant gratification, which company wants to support a system that needs you to asnwer 35 questions (MediaUnound) as opposed to only one (Amazon, CDNow etc.).
Manu, this was hardly an anti-corporate rant. And I am quite passionate about Recommender Systems. There is a lot of promise of Recommender Systems that has not reached people. Yet. I live in hope that it will. I do not exaggerate when I say that recommender systems helped me revive my interest in music, find new tastes, remind me of music I used to like. It was more powerful information filtering that I have ever experienced.