Personalized Recommendations Are Still Hit Or (Mostly) Miss
from the thumbs-up,-thumbs-down dept
Just as Google is throwing its hat into the personalized news recommendation ring, the NY Times has an article looking at a variety of recommendation systems and basically concludes that it’s tough to have recommendations that actually are all that useful. Some systems, such as those used by Amazon, Netflix and iTunes (all using some form of collaborative filtering) have pretty strong reputations — but even then, it’s amazing just how many of the recommendations people see that simply don’t match them at all. The real issue is that automated personalization is an extremely hard problem to solve algorithmically, and it still seems like the “best” ones are only picking out relevant info only a small fraction of the time.
Comments on “Personalized Recommendations Are Still Hit Or (Mostly) Miss”
My TiVo thinks I?m gay
My TiVo thinks I?m gay… and the NY Times now only shows me the comics and horoscope sections.
In all honesty though, pushing people to a narrow field of information often makes those people bored… and they will often go elsewhere for “variety”.
Just because I enjoy watching movies about gladiators, prison fights and gangs, doesn’t mean that I like movies about men – it means that I like action movies and could care less about plot and story line. I prefer a good James Bond, Stallone and Arnold movie when they are on – but give me an entire channel of women, and im hooked! (i don’t mean Home & Garden network or TLC)
I think the same could be said about all these other “we recommend” systems – they really don’t know what individuals want. Only the individual knows this.
Not all that useful?
basically concludes that it’s tough to have recommendations that actually are all that useful.
How did you get that from the article? From my read, the article says exactly the opposite.
At the beginning, she says, “Companies are finding that getting those personalized recommendations right – or even close – can mean significantly higher sales.”
On Netflix, she says, “roughly two-thirds of the films rented were recommended to subscribers by the site” and that Netflix “credits the system’s ability to make automated yet accurate recommendations as a major factor in its growth from 600,000 subscribers in 2002 to nearly 4 million today.”
Re: Not all that useful?
I like the way Netflix does it. Their personal “recommendations” are always way off base, but their “If you added this movie to the queue, you might like these as well” hits dead-on about every single time.
Re: Not all that useful?
My amazon account reccommends books like Bridget Jones Diary 2 and Victorian Literature because I’ve bought far more books for my classes as an English major than I have for myself.
My roommate last night was looking for Freshmen, the comic by Seth Green. Well there’s a magazine, Fresh Men, that came up in his search so now amazon thinks he likes young men freshly 18,
If they let you click a box that says I’m not really interested in other things like this they might end up with better recommendations.
Pandora
I encourage any music fans to check out http://www.pandora.com for an idea of how a well-designed “you might like this too” index can work.
More targeted recommendations
Netflix’s main system seems pretty hit or miss…for some reason it’s determined to keep recommending Dragonball Z and wrestling titles to me, despite the fact that I constantly rate them “not interested” and I hate anime.
But I love their “Friends” feature! I have pretty eclectic tastes, and several friends also into weird and obscure things, and seeing what they watch and rate is far more useful. If I know what somebody likes and dislikes, their ratings mean far much more to me. I’ve discovered some neat films that way, and also turned several people onto hidden gems and classics that way. The only downside is my queue keeps ballooning out of control as a result.
I’d like to see more of that. Amazon could say something like “Brian loved this book”, and since I know his tastes it’s a far more useful recommendation to me than “People that bought this also bought that.”
Re: Netflix recommendations are a mess
I have rated 1000 movies from every genre, I have seen a few Kurosawa flics and coupe of samurai dramas, yet my recommendations are flooded with mnga/anime/cartoon, I have never rented one. I have given up on their personalization, it has no way of predicting that I like to watch good movies regardless of genre. What I consider good no personalization system can predict…