Why the heck does Netflix think I'll like ____?

Tom Selleck and Pierce Brosnan? I’ve gotten recommendations for 4 of each actor’s movies, and I’ve never rented or shown interest in anything involving either of them.

Post other baffling recommendations a computer program has made for you.

There’s always the classic “my Tivo thinks I’m gay”.

I actually emailed Amazon, once, asking why, since I’d bought nothing from their music department except obscure pieces by Sibelius, they were recommending Alice Cooper … pointed out there was a bit of a difference between (for example) Luonnotar (op. 70 for orchestra, piano and solo soprano) and Mr. Cooper’s little ditty “Raped and Freezing”.

I didn’t get an informative answer back.

My company writes the software that makes these recommendations, but not for Amazon or Netflix. There can be some subtle connections made by movies that are grouped as similar genres in the industry standard data. For some applications, we provided information to the end used about why certain items are recommended, but not in all cases.

I don’t know how Amazon or Netflix has implemented their algorithms, but there are also boosts that can be applied if they want to move certain slow moving titles or ones where they have extra inventory.

I think Netflix has admitted in the past that their recommendations need some work, but that was a few years back.

Yup, they had a job offer on their website a while back to anyone who can create a better algorithm. It must be a hell of a lot harder than it sounds, because it seems years later no one has taken them up on it. There are so many reasons to like movies you can’t quantify. I like Bruce Willis but hated Armageddon. I like war movies but hated Thin Red Line. I like Joe vs. the Volcano…I still don’t even know why.

I just checked and they had no recommendations for me. Based on their past, I don’t mind, since they are usually so far out there that I wondered if they had my account confused with someone else’s.

I started a thread asking exactly the same question a while back, and the consensus seemed to be that they recommend movies that were enjoyed by people who enjoyed the movies you have. So if you liked Ishtar, and someone else liked Ishtar and Armageddon, Netflix would recommend Armageddon for you. So the recommendation wouldn’t necessarily have anything to do with Armageddon itself. I don’t know if that’s actually true, but it makes sense.

My previous thread.

Yeah, that would be my guess too. Could be that some couple has a Netflix account and he gets movies and she gets movies and they don’t really have the same tastes…but it sort of mucks up the recommendation scheme when it comes to “other people who rented this movie also liked…”

Sometimes I’ll get movies that I probably wouldn’t like but they’ll be for my mom or my boyfriend. Throws off my recommendations and probably all others.

NetFlix is offering a million dollars if you can beat their system by better than 10%. The contest runs into 2011. Some of my research colleagues were working on it but I don’t know how far they got. It is definitely a hard problem: search Google Scholar for “collaborative filtering” and “recommendation systems” to read some of the current research being done on the problem in general.

My experience is that NetFlix and Amazon will recommend stuff that I already know I like, but it’s very very rare that I get a recommendation that’s surprising, i.e. something that I didn’t think I would like or didn’t know about but ended up liking. I don’t get many really awful recommendations, but I do get a lot that don’t interest me.

Besides being annoyed by ridiculous recommendations, most of the ones I get I have already seen.

It doesn’t seem that it would be that difficult to scan my order list and not put the ones I’ve already seen up again.

Probably is too difficult. :smiley:

Blockbuster was no better. I’d rented nothing (nothing I say) but action and sci-fi. Bourne, Split Second, War of the Worlds, things like that, and my number one recommendation was a foreign language romance (and not one of the good A/O ratings).

I figured they were worried about me and trying to expand my horizons, or something.

FlyingDragonFan is correct. The recommendations are based on what other members select that had out the same movie. In fact, it even states that on their website

followed by a list of 3 recently viewed movies. The recommendations come from other members who watched those same movies and is based on what else they watched and rated.

Netflix had been shoving “Donnie Darko” on me for ages. “Dangit,” I’d say, “Why do you keep telling me to watch this movie? It looks gawdawful.”

Finally, I gave up and rented it.

And you know what? It was pretty good.

Maybe you should trust the computer and rent a Tom Selleck movie.

The part I can’t figure out is why Amazon, which knows full well I was browsing Elmo and Thomas the Tank Engine videos for my son back when he liked that kind of thing, is still recommending them to me five years later.

Someone needs to tell the system that kids grow older, and their interests change. Someone who was interested in baby toys in 2001 just might be in the market for a little kid toy in 2007, you know?

That’s bizare FisherQueen, they must be pushing that movie because I just rented Donny Darko because it was recommended. I can see the connection because I like some pretty weird movies. I wasn’t a fan of Donny Darko though. Without giving away the plot, I don’t like movies that end that way.

I usually pick things out of those recommendations. Sometimes I’m really suprised at how good they are. My recent success was 12:01, sort of a SciFi Groundhog Day. It wasn’t high art but it was thoroughly entertaining.

This one I don’t get either! If I’ve rated it previously, I’m pretty that means I’ve seen it, excluding the ‘Not Interested’ rating. If I’ve rented it previously I really shouldn’t be recommended it.

As it is, I don’t know why I bother. I have 40 movies on my list and with the one disc at a time plan, I see about 4 a month.

Our code has filters for things on rental queues and things already seen/rated. There may be times that you want to recommend something that has been rented before. The business rules can get tricky depending on what behavior you are trying to encourage.

Recos are tricky, since you don’t want to recommend the top 5 scoring movies if those are all very similar. Putting a range of movies in front of the user is more likely to get someone to add things to their queue then a block of 5 that resemble each other very closely.

Well, right now on Netflix, I have the following recommendation:


psst You know you can remove items from your browsing and purchasing history so that it doesn’t recommend items based on them any more, right?

It took me a couple of hours to remove all the junk that accumulated from purchases for friends, children of friends, relatives, trying out shows that we’ve never gotten out here in Australia etc. but Amazon is finally clear for me on recommendations now.

But I agree that an automated system for clearing out purchases would be nice.