Since I am a Lord of the Rings nerd, people assume I love Harry Potter. Nope. After the first book it just got too repetitive, derivative, and just plain boring to me.
Since I enjoy science fiction, people assume I love Star Wars. Nope saw the first as a grad student. wrong age, I think.
Since I enjoy some YA books, people assume I love Hunger Games. Nope. Can’t stand books written in the present tense.
and I’m not impossible to please, seriously.
Many sites, such as Amazon will give you suggestions based on past likes. Do you guys ever find these helpful? They can’t ever know what variables your likes & dislikes are based on, can they.
Thoughts?
When I started trying to put together a complete list of movies I had seen in IMDb, I found their recommendations very helpful in spurring my memories.
People like to take shots at Netflix’s recommendations, but they’ve always worked pretty well for me. Sure, there’s the occasional WTF suggestion, but when I add a movie to my queue, a lot of times, the recommendations will contain at least one movie that makes me say, “Oh yeah, I’ve been meaning to watch that.”
That’s how the suggestions are helpful to me: they remind me of titles that I’m already somewhat familiar with but have somehow slipped through the cracks of my memory.
I’ve said here in the past that Amazon’s recommendations might just as well be random picks from the database for all that they’re relevant. Not too long ago I bought a book they recommended based on my love for Sherlock Holmes. It was horrible and really, the only two things that they had in common were being set in historical England and *labeled *mystery.
As someone pointed out, “might” is the operative word. I find amazon’s suggestions very helpful and I like that they do that. Maybe you ARE hard to please and just don’t know it?
These people are idiots who are comparing things on the basis of very shallow attributes. Of course they aren’t useful suggestions. You wouldn’t hand someone a block of American cheese just because they said they like oranges and they’re both orange.
A problem with Amazon’s engines is that when you buy things, you don’t always buy them for yourself. This tends to confuse the engine about what you enjoy. I bought my dad a book on digging for clams, that doesn’t mean I’m interested in more books about shellfish.
I find Netflix to have an excellent recommendations engine, one thing I also notice is they have fairly deep level of assigning attributes. You’ll see lists like “quirky, bittersweet urban comedies” or “historically accurate war dramas” as opposed to “comedies” or “war movies.” However you have to rate a lot lot lot lot lot of stuff before it makes good suggestions.
I am the worst candidate for these type of things. I like the best of everything. Sure, every once in a while these systems work but for the most part.
I would rather have the best of a completely different genre than the C level contributions of a film I liked given genre.
Yeah maybe. Here, I was recommended: How to be a Super Hot Woman: 339 Tips to Make Every Man Fall in Love with You and Every Woman Envy You [Paperback]
Because:
*Recommended because you purchased Hyland’s - Leg Cramps W/ Quinine, 100 tablets *
I’m a man BTW and am really not at all interested in being a woman.
This one was good too: Pro ASP.NET 3.5 in C# 2008: Includes Silverlight 2, Third Edition [Paperback] recommended to you because you purchased: Pro ASP.NET 3.5 in C# 2008: Includes Silverlight 2, Third Edition [Paperback]
This shouldn’t be a problem - you can mark anything you’ve bought as a gift, which tells the engine essentially what you said “Just because I got Dad a book about Clam Digging doesn’t mean I want more books about shellfish”.
OTOH, the recommendation system is really, really bad at video games. You’d think they could AT LEAST reason that if you’ve never bought a game for a system, that that might be because you don’t OWN that system, so they could stop recommending me Nintendo 3DS titles. PLUS, the matching mostly seems to be “You liked a video game that sold well. Perhaps you’d like some other games that are selling well?”
The one that absolutely cracked me up was when they informed me that I liked “political dramas involving revenge” because I had rented The Merchant of Venice and Measure for Measure. I think those two movies might have something slightly more obvious in common
And also: you bought a book from *this *author, so here are a dozen others by the *same *author. Gee, it *never *occurred to me to check out the other books by him.
And another: You just bought a dish washer; here are 20 other brands that are similar. How many dish washers do you think I need?
I find it useful maybe 35% of the time, and annoying 65% of the time. The mistaken recommendations don’t bother me quite so much, as I can click on the “Fix this recommendation” and change it.
But it is so very difficult to make lovely serendipitous discoveries on Amazon or other web-based stores. If you do general browsing, there is so much dreck to wad through, it’s hard to find the gems. That’s what I like about browsing in a good indie bookstore – the buyers have already waded through dreck in the catalogs, and made intelligent selections for the store. And there, I make the lovely unexpected discoveries.
I find Amazon’s recommendations to be useful about 30% of the time, which isn’t bad. And most of the time when they aren’t useful it’s because of some obscure actor appearing in one movie also appears in 35 other movies. “No, buying the Blade box-set doesn’t mean I want to watch every Ryan Reynolds movie ever made, thanks.” Of course I can also go in and edit the way they guess at the recommendations so that won’t happen as often, refining it down a bit more accurately.
Anyway, it works okay for me. Buying a Dara O’Briain DVD means I get recommended a bunch of other UK Stand Up Comedian DVDs, many of which I end up finding very entertaining.
I used to work in the recommendation algorithm business, although it was a few years ago and technology and approaches have changed somewhat since then. But the algorithms are all data driven, often based on statistical correlations between purchases by the masses. This works well when there’s lots of data - and most people who like Lord of the Rings do like Harry Potter so that recommendation is probably a good one. These algorithms often break down when there’s not enough data to make a reasonable correlation. If only one person bought the book that you bought, the algorithm might recommend his purchases and the chances that they are good for you are pretty slim. A good algorithm should have a minimum threshold of data points before valuing a recommendation, but at the thresholds those recos are going to be less valuable.
Other parts of the algorithm depend on the classification and categorization of the data set. If the data is rich you have lots of characteristics to work with; if the data is poor then all the comedies are lumped together. Concrete characteristics (like actors, directors, genres) get more weight than things like “light” or “dark” which may not be available across the entire data set.
Another piece of the algorithm regulates how many familiar versus exploratory suggestions are presented. You don’t want to always recommend things the user is likely to be familiar with; that gets boring. You mix it up with some educated guesses that the user might find interesting or catch his eye, but won’t be recognized immediately. That is responsible for some occasional oddballs.
It’s a complex process, and one that almost by definition can’t work for everyone equally well.
Actually, we recently got access here at the library to a new reader’s advisory tool (a way of recommending books to people who read things that, for example, you have no personal knowledge of). The first thing I did was check it to see if it would recommend other books by the same author like so many of them do.
It did not. I tried a different author, still not. A different author, still not.
Finally I wrote it off as a new resource with a fairly shallow database that hopefully would deepen over time and went looking through its documentation. That’s where I found a statement that it explicitly excludes other works by the same author from coming back in the results, on the grounds that duh.
I find these recommendations humorous because of one pretty big gap in the system: the pool of “stuff you like” is polluted with other family members’ “stuff they like but you don’t really like”
In Amazon, you can knock out books and videos you bought as gifts, but you need to know this and you need to actually care. I don’t know if you can do this in Netflix.
A couple of weeks ago I got a nice email from Netflix saying something like “Dear Member, we thought you would like to know that the latest season of Gossip Girl is now available”
Gossip Girl?
I simply forwarded that to my teenage daughter, with a note that “I think this one is for you” She was pleased.
(ETA: I just noticed that Hello Again alluded to this too.)
I’ve found Amazon’s recommendations to be somewhat useful. Sometimes the suggestions are logical, but not applicable. For instance, I loved Disney’s Mulan. However, I am not interested in any of the Disney Princesses toys or lifestyle accessories for little girls. And I rated several games for the SNES and the PSX/PS2, but that doesn’t mean that I want games for the Wii or DS. I’d love to be able to tell Amazon “I don’t have or want a Wii, so quit suggesting games for this platform”. Same thing with Blu-Ray, don’t have one, don’t want one, I’ll probably break down and get one eventually but until then I just automatically check “Not Interested” in all the suggestions for Blu-Ray movies. I’d also like a way to tell Amazon that I am not ever going to buy certain authors. For a while, I was being shown every single book Robert Jordan ever wrote, in hardback, paperback, and used. Yes, I like epic fantasy. I don’t like Robert Jordan, though.
However, I’ve had some really, really great suggestions, too. I find that the suggestions tend to work better on nonfiction than on fiction, for some reason, though I’ve had some success with fiction writers as well. I probably wouldn’t have picked up John Scalzi’s books except that Amazon kept recommending him to me.
I visited friends who insisted that I would love The Big Bang Theory because of my love for other nerdy things. They even made me watch an episode, which just served to confirm my opinion that BBT has never left any nerd cliche unused.
