We just rejoined Netflix after several years away from it. While recreating preferences and ratings on a fresh account, we noticed something surprising: Netflix doesn’t allow a user to judge things as “just ok.” Take a look at the tool-tip explanations for their five-star ratings:
Netflix’s recommendation system is generally considered its most valuable asset, so much so that the company sponsored $1 million prize to improve it. Yet, I wonder how much expectation factors into predictions of user ratings.
In particular, I find many films to be “pretty good” or “just OK.” These are the movies I didn’t mind watching but don’t need to watch ever again. They are the films that I “enjoyed” in terms of their delivery of a satisfactory movie-watching experience, but not necessarily for their artistic or expressive merit. How do I rate such movies?
My instinct tells me that I should rate them 3 stars out of 5, right in the middle of the range. But Netflix tells me that 3 stars means “liked it,” while 2 stars is already “didn’t like it.” I need something in-between.
I suppose one might counter that the textual descriptions of the star ratings system have nothing to do with the statistical analysis of their results or the prediction algorithms that manipulate such data into recommendations. But that’s not true if consistency matters, either within a single user’s ratings or across users. If I’m using 3 stars to mean “just OK” at some times and “liked it” at others, then I’m poisoning the statistical well. Likewise, if I’m using 3 stars to mean “pretty good” all of the time but another user is using it for “liked it” all of the time, then recommendations made across our viewing patterns could be inaccurate.
Technical matters aside, I find it odd that Netflix would think I could only love, like, or hate films. I re-subscribed partly to take advantage of streamed rentals on Xbox Live, since it doesn’t require waiting for discs in the mail. In such cases, it’s not always required that I pass aesthetic judgement on the films I watch. It might be enough that their images and sounds simply reach my retinas and eardrums, successfully elapsing 90-120 minutes.
In this respect, the idea of an instant queue is already odd. But the notion that I should feel obligated to like, love, or hate something in such a case misconstrues a major mode of film-watching, and perhaps of being itself.
What Netflix doesn’t support is a sense of “meh,” that term for indifference made meme by the Simpsons a decade ago. The Italians have their own version, “boh” (or sometimes “buh”), which means roughly the same thing, but has a rounder, more definitive sound to it.
Consider the episode of the Simpsons that popularized “meh” (season 12’s Hungry Hungry Homer). The family visits Blockoland, a theme park of anonymous slabs.
Homer: (after watching Blockoland commercial) All right, kids … who wants to go … to … Blockoland?
Bart and Lisa: Meh.
Homer: But the commercial gave me the impression that …
Bart: We said meh.
Lisa: M-E-H. Meh.
Many, from Benjamin Zimmer in his essay on the word to the Collins English Dictionary in their entry for it, misunderstand “meh” as a term of disapproval. Zimmer and others have also tried to connect the term to the Yiddish slang “feh,” an interjection of mild disgust usually accompanied by a dismissive hand-gesture. But all of these accounts miss the truly ambiguous wonder into which “meh” has evolved.
While “meh” may seem to imply disapproval or disinterest, it is actually a sophisticated response of veneration, a celebration of something’s ability to be a part of the anonymous formlessness philosopher Jean-Luc Nancy calls “whatever.” Bodies for Nancy spread outward rather than cohering, like a bag of gummi bears on a summer afternoon. Here we can see a relation to Heidegger too, for whom being withdraws, and when pushed even further, to Graham Harman’s understanding that all objects withdraw from relation.
It is convenient that “whatever,” in English anyway, also denotes the kind of apathy we associate with “meh.” But meh is more like Nancy’s whatever than like ours: rather than isolating a single example as wanting, it underscores the indelible anythingness of being.
Meh is an expression of impassivity, one that signals neither apathy nor boredom, but something that sits on the razor’s edge between the two. Meh is nonchalance instead of rejection, meh is apassion instead of dispassion. Meh is an affectionate detachment. Meh is moving on. Meh is a zest of blandness. Meh is warm but not luke. Meh deadpans. Meh is the water on the rock, the wind on the sail, the tube on the toothpaste.
When thought of this way, to “meh” is to express the sublime, but not in Kant’s sense of a dominion over us. This sublimity is one of a level playing field, of a flat ontology, to use Levi Bryant’s term. It is a mundane sublime, one that is better characterized by Stephen Shore than by Ansel Adams. Following Harman, meh reminds us of the world’s insatiable ability to be so much and so many, all at once, whether as riverstones, as fur, as Fiat 500s, or as romantic comedies. It is a magnificent blandness that breaks star rating systems entirely. O Netflix, let me slouch and mutter it, without exclamation point, “meh.”!
Comments
Nirmal Patel
The lack of a neutral response is fairly common in survey based interaction research. Many Likert scales have an even range so the user has to fall on one side of the like/dislike range. When given the option to not have a decisive opinion many people will take it.
Ian Bogost
Nirmal, if I’m not mistaken, there’s also evidence that people tend not to use the bottom end of rating scales (the 1 and 2 star, in this case), but to adjust toward average. I can’t remember where I read this.
It’s a shame in both cases, anyway, because the neutral response is not necessarily neutral, nor a cop-out. It may suggest something else entirely, a reaction that sits outside of approval/disapproval.
Nirmal Patel
Yep, you’re exactly right. Everytime I’ve used a Likert scale, someone always tells me they want to pick the very middle and they always have very specific reasons (i.e. not a cop-out).
Frank Lantz
I think rating/recommendation science is fascinating.
Many recommendation researchers feel that accuracy isn’t the most important problem, and that the Netflix prize over-emphasizes accuracy at the cost of features like novelty and transparency. http://www.grouplens.org/papers/pdf/mcnee-chi06-acc.pdf
The meh problem is sort of a flipside to the Napolean Dynamite problem http://www.nytimes.com/2008/11/23/magazine/23Netflix-t.html?pagewanted=all. We would like to teach our machinery how to love and hate, and also how to shrug.
But I am willing to bet that we are not going to solve these problems, no matter how finely we slice our thermometers and tune our algorithms.
Because movies are art, even the bad ones. And artworks themselves are machines whose job is to confound predictions. Even as we apply our taste functions to a movie the movie reaches into our taste functions and modifies them, like a recombinant chemical cocktail that reprograms the tongue. This is the G̦dellian incompleteness of film Рmovies will always have the capacity to disappear into the irresolvable infinity between two numbers and the unnamable emptiness between two words. And we will continue to know what machinery cannot.
Tom Clancy
I run into the same thing with iTunes: I really have to hate a song to give it a 1. Otherwise something gets at least a 2 just because I feel bad (which is crazy). So I wind up wishing there were 10 stars, but then I realize I’d do the same stupid thing with that, dividing it up in Zeno’s paradox of “What does 7.25 *mean*?”.
I feel the same way as you do about the Netflix scale. I wish they wouldn’t bother to tell you what the star ratings should mean for you because I find myself inventing a system for most every rating tool I use. Netflix, Amazon, iTunes all use 5-star ratings, but I use each in a different way.
Sometimes I feel stupid and OCD for doing that, but then sometimes it occurs to me the 5 star rating concept was not passed down to Moses from God either.
Greg
See Sianne Ngai’s book Ugly Feelings (Harvard UP, 2005). She writes of ‘stuplimity’ (dedicates a chapter to it), and its noncathartic aesthetic of boredom and low-grade aesthetic awe (the synthesis of awe and what refuses awe). Stalled or suspended agency. Kant’s apatheia. Neutral, unqualified, ‘open.’
Abbey
Unfortunately, I feel no compunction at all about doling out 1s with great liberality. It’s the 5 that I just can’t bring myself to click. Except maybe for Labyrinth. The Goblin King can have a 5.
Unfortunately, the more movies I rate, the fewer Netflix is willing to recommend. The “Movies You’ll Love” section has shrunk to almost nothing, while I frantically rate more in the hopes of new and creative suggestions.
So I click the little box that says “Rate more [genre] so that we can give you more suggestions”. And I’m faced with endless seasons of Friends to rate, which presents two problems: first, I haven’t the foggiest idea what differences there may have been between seasons 5, 12, or 28, and second, I really do feel entirely “meh” about them. I’ll watch an episode of Friends if it’s on in front of me. I’ll laugh at the appropriate moments — sometimes a slight giggle, sometimes a gasping-for-breath guffaw — but I’m never going to think, “You know what I want to get from Netflix next week? Hours and hours of Friends!” Ever. It’s just not going to happen. So I scroll through, rating season after season (after season) of friends, only to be faced with seasons and seasons of something I hate, or something I don’t care about at all, or something else that I like well enough but would never bother to rent…
So Netflix punishes me for my negativity.
“You? You’re never gonna love anything.”
Mattcc
I’ve actually written to Netflix to complain about this very thing. I explained that there are so many films I neither like nor dislike, OR that I simultaneously like AND dislike for different reasons (the ending was a cop-out, etc), so that I can chose neither side of the fence and would like to sit on top please.
I don’t remember receiving a response.
Ian Bogost
A clarification: over at his blog, Levi correctly notes that “flat ontology” is a term borrowed from DeLanda, although the latter uses it differently. Graham Harman also mentions that others (e.g. Roy Bhaskar) have deployed it in even different understandings. In any case, the sense I wanted to highlight is the one Bryant has adopted.