You Aren’t Who You Hang Out With
Every new app you try these days wants to know who your friends are. It’s easy to understand why. On the marketing side, it’s to encourage users to evangelize the app amongst their friends. On the user experience side, however, it’s to help users consume more relevant content.
Here’s are a few examples:
- Upon signing up for Rdio and connecting your Facebook account, you are shown music your friends are listening to.
- Upon installing Oink and connecting your Twitter account, you are shown food and other items your friends have sampled.
- Upon checking your Facebook news feed, you are shown status updates from friends reacting to movies they’ve just seen.
While this sort of content tailoring provides value, I often find myself uninterested in it. The reason is that although in many cases my friends are similar to me, my taste in things like music, movies, and food do not map to my friends’. The taste correlation between friends may be greater than between two random strangers, but it’s still not very high in most cases.
There’s a better way to expose people to new experiences and I think we’ll start to see more of it in the future. It may already have a name, but I’ll call it “phantom friending”.
To illustrate phantom friending, imagine you want to watch a movie tonight and you need a recommendation. Now imagine you have these two options:
- Calling your best friend, asking them what good movies they’ve seen recently, and picking one of them.
- Consulting a list of preferred, recent movies put together by someone across the country who you don’t know but who has in the past indicated that they hate a lot of the same movies you hate and love a lot of the same movies you love.
I hold that in almost every case, the second option will provide a better result. Even if you were able to poll 5, 10, or 20 friends, a well-picked phantom friend would produce a better result. That is because the phantom friend doesn’t represent someone you like to socialize with — as your real friends do — but rather someone who watches movies the same way you do. They have your same tolerance for violence, same appreciation for special effects, and same patience for heavy dialogue. In other words, they may be unlike you in every other way, but their brain consumes movies the same way yours does.
The phantom friend concept works better for some subjects than others. It would seem to work well for movies, food, and music. It may work less well for TV shows, because a big part of TV shows is discussing them week after week with our friends. The same goes for clothing. We often wear similar clothing as our friends in order to fit in better.
For the many situations where phantom friends are better influencers on us, I’d love to see more apps and services geared towards this type of discovery. One example I’ve always wanted is a “Movie Critic Dating Game”. I rarely read movie reviews because I haven’t identified a movie critic who is a lot like me. Here’s how it would work:
- I am presented with a list of 20 movies.
- I rate each movie with a thumbs up, thumbs sideways, or thumbs down.
- The app finds me the national movie critic who has rated the 20 films most similarly to how I have rated them.
- I then begin reading the critic’s reviews each week and choose new movies to watch accordingly.
Interestingly, the above scenario works almost as well if the system can find someone with the exact opposite tastes as me. If I can find the person who I disagree with the most, I can just always do the opposite of what they suggest (the “Costanza strategy”). Furthermore, even if you extended the questionnaire to 200 movies, there is someone in the world (although perhaps not a professional movie critic) who answered all 200 the same way you did.
Undoubtedly I am not the first to think of this concept, but given that it doesn’t seem computationally ferocious to do, I’m surprised we haven’t seen more of it. Hunch seemed like it was after a similar result, but it always seemed too impersonal to me. I don’t want a computer telling me what people similar to me like. I want a computer matching me up with someone and then letting me know what else they like. There is a difference there.
I can imagine a world in which I have a movie sensei, a restaurant sensei, a music sensei, and a bunch of other senseis. I may eventually know them by name or I may not, but it would be a fun set of relationships to have.
Good point. I stumbled upon a shopping recommendation site (think Pinterest for goods) called Svpply today. It’s based around the friend model, too. I found myself doing just what you did, though. I started adding people I follow on Twitter who I really admire and find myself aligned with. I don’t know them in person, but their taste is really valuable to me.
A couple disparate points:
• “Friends” seem to come in a few flavors these days, particularly if you’re the type of person who would read this blog. 1. Twitter friends: typically not real-life friends but more like professional acquaintances, random funny people, various celebrities, etc. 2. Facebook friends: family, actual friends, old friends, long lost former classmates. Of those two groups, I’m not sure I’d outsource my content recommendation to either one of them. While it’s interesting (and maybe even informative) to see which content they consume, it’s not doing much to find content that’s relevant to me. There’s definitely a third class of friends out there (phantom/taste friends) that no one’s yet found an effective way to corral.
• Lately, I’ve found myself downloading new apps—Oink, Stamped, the new Path, etc.—and following designer/developer friends as a means of determining how they use the app. I’m generally not dedicated to using the app, I simply want to see the UI and the usage patterns…maybe even how they break it (see DickProofing™). As a member of the class of people who build these things, I’m finding it more and more difficult to separate “Jay the user” from “Jay the designer.”
A weird thing happened to me. I wasn’t logged in to Facebook. I signed up to Twitter. When it then recommended friends for me to follow, they included many I knew from Flickr and Facebook! But… how? Neither service was running at the time. So how did Twitter know about these friends?
I can only assume that either a) data is being shared between services without our knowledge, or b) maybe it checked a cookie on my computer or something left over from my last Facebook session, assuming such a cookie is used in that way.
The truth could be a third option I haven’t thought of yet – maybe it just did a search against my name? Yet if you type my name into Google, you get hundreds of people with the same name. Perhaps geolocation played a part too.
Scary.
I followed you on twitter because when I enter IRC/elsewhere as MikeD I am typically asked if I am you. So far it seems we are fairly similar in attitude regarding technologies.
I’m not the MikeD from the Beastie Boys either. :)
Awesome notes Mike. I completely agree with your phantom friending examples. In fact, building through them as we speak. I think it’s important to allow people you do and do not know to help your curate. Most importantly not limiting the people who don’t have the same tastes as you if the technology is making that correlation. Often if the technology assumes it, we’ve found it’s not always correct.
The process of deciding if someone has similar tastes as you is often not 100% yes or 100% no, it’s a mixed bag and lies somewhere between the 40% to 60% mark. So we’ve declared; you decide based on your ask not system level before it reaches you. Ideally this scenario of others without your tastes coordinating with your ask could let you develop taste and not let it stay stagnant. But this is still a theory.
I’ve been using Criticker to help me find movies I may like. What the algorithm essentially does is compare your taste to similar users and then averages their ratings for movies you haven’t seen to generate a score/likelihood that you’ll enjoy seeing it.
It’s not a perfect system. Seeding your initial ratings database can take a while (depending on how thorough you want to be). Composite scores can be completely off at times (it’s very difficult to find someone with exactly the same taste as you).
I’ve stumbled across quite a few gems I never would have otherwise, though.
You know you like my taste in movies. What was the last one we saw, Contagion? That was serviceable!
Malibu’s Most Wanted is an excellent way to whittle down your list of matches–there are only 4 top critics on RT who think it is “fresh.” Good luck!
Mike: More importantly, does anything mistake you for the great Mike Doughty?
Patrick: Regarding the 40-60% thing, isn’t that the case if you are talking about someone’s overall tastes though? My thing is that I don’t want to ever talk about overall tastes. I want to talk about specific tastes. Like taste in movies, for instance. I feel like there are people in the world who have almost 100% the same taste in movies as I do. The problem is when you try to match those same people to other aspects of my life, like eating.
Peng: Thanks. Going to give Criticker a try!
Brade: Totally. Very polarizingly awesome movie. :)
Perhaps run some semantic analysis to gauge your friends queue compared to yours to find out which “friend” could be a phantom friend and start there as well as tapping into other feeds. This is similar to the DJ’ing at blip.fm. You follow DJ’s that have similar taste in music not just my friends.
Some of this could also be initiated by a 25 movie “hot or not” quiz. Lots of concepts in here.
j
Pearson Correlation FTW.
iCheckMovies.com does a pretty damn good job of matching you up with your phantom friends for movies. They call them “neighbors”.
And then they give you lots of different ways to slice up the data: movies you’ve both seen, movies they have seen and you have not, movies you have seen and they have not…
Funny, but I used to use the early music sharing programs the same way…
I’d search for a song, get back a bunch of hits of people who had the song, and some of the programs would allow me to click on the user name and see their entire list of songs.
It was really exciting b/c I made the assumption that b/c they liked some of the same stuff I liked, I should take a look at other music that they liked. I discovered a lot of great music that way that I never would have otherwise…
And then the music stopped…
Dan: You’re totally right! That’s exactly what Napster did! Find some rando who likes the same stuff you do, and then look at the rest of their stuff. That seemed to work pretty well… interesting that things have moved away from that.
[…] You Aren’t Who You Hang Out With | Mike Industries […]
Netflix used to do exactly this with their movies reviews which they removed for some reason.
When you would read a movie’s reviews there was a percentage that showed how similar the reviewer’s tastes were to yours based on the movies they had previously reviewed. I found it very useful.
Christian, Netflix removed it because they will always make the wrong decision on what features to keep and what to remove. I’m only kind of joking.
According to LifeHacker (http://lifehacker.com/5884202/five-best-movie-recommendation-services), it looks like Criticker (http://www.criticker.com/) is what you’re looking for.
I just saw this, so I have no idea how good it works, but I figured it’s worth trying out.
Kyle: I really wish Criticker wasn’t so poorly executed. Asking me to rate a film 0-100 is already too taxing. I’d be more likely to rate 100 or more films if I could just say “Dislike, Meh, Like”. Even A through F would be better. 0-100 requires way too much thought and way too much fear that you didn’t answer the question accurately.