You May Also Like

by Tom Vanderbilt

  All this does not mean that ratings, having been pushed in a certain positive direction, always rise. In fact, on a site like Amazon, while “sequential bias” patterns have been found, there is a general tendency for book ratings to grow more negative over time. “The more ratings amassed on a product,” one study noted, “the lower the ratings will be.” What distinguishes Amazon from the one-click liking or disliking mechanisms seen in Aral’s experiment is the higher cost of expression: You cannot just say how much you like or dislike something; you have to give some explanation as to why.

  This seems to change behavior. As the HP Labs researchers Fang Wu and Bernardo Huberman found in a study of Amazon reviewers, in contrast to the “herding and polarization” effects seen at the Digg-style sites, Amazon reviewers seem to react to previous “extreme” raters. Someone rating on the heels of a one-star review may feel compelled to “balance it out” with a three-star, when in reality he was thinking of leaving a two-star review. This reaction to extremes can lead to an overall “softening” of opinion over time.

  One reason, they suspect, is an inherent desire to stand out from the crowd, to actually affect the result or inflate one’s sense of self-worth. “What is the point of leaving another 5-star review,” Wu and Huberman ask, “when one hundred people have already done so?” Rationally, there is none, just as, in the “voter’s paradox,” there is little rational sense in voting in elections where one’s individual vote will not affect the outcome (although, unlike with voting, there is evidence that recent reviews do affect sales). So the people who leave opinions, after time, tend to be those who disagree with previous opinions.

  It is easy to imagine several stages in the evolution of a book’s ratings life on Amazon. The earliest reviews tend to come from people who are most interested in the book (not to mention an author’s friends and relatives, if not the author herself) and who are most likely to like it.

  Taste is self-selection writ large. But once an author’s fans and other motivated customers have weighed in, over time a book might attract a wider audience with “weaker preferences,” as the researchers David Godes and José Silva suggest. Whether they are more clear-eyed and objective critics, or they do not “get” a book the way early reviewers did, their opinion begins to diverge. With many books, a pronounced “undershooting dynamic” kicks in: a period in which reviews are even lower than the eventual lower average, as readers, perhaps swayed by the previous “positive review bias,” make essentially mistaken purchases. Then they weigh in, in what might be called the “don’t-believe-the-hype effect.” So begins a feedback loop. “The more reviews there are,” Godes and Silva suggest, “the lower the quality of the information available, which leads to worse decisions which leads to lower ratings.” It is not uncommon to find late, fairly flummoxed one-star reviews of only a sentence or two: “I just didn’t like it.”

  As more reviews are posted, people spend less time talking about the contents—because so many other people already have—than about the content of the other reviews. When a review mentions a previous review, it is more likely to be negative. Context takes over.

  Which leads us back to Aral. How can you actually tell if a review has been influenced by a previous review, or whether it is simply homophily: correlated group preference, or, more simply, the idea that “birds of a feather flock together”? He gives an example from the sociologist Max Weber: If you see a group of people in a field open umbrellas as it begins to drizzle, you would not say they are influencing one another. They are merely people reacting to endogenous conditions, with a correlated group preference to not get wet. If a person opening an umbrella when there was no rain could get others to do so, that would be social influence.

  —

  Teasing out what is going on in online reviews, as with taste generally, is a messy affair, one that Aral has grappled with for a decade. Do people in my Subaru-heavy neighborhood buy Subarus because they see lots of other Subarus, or do Subaru buyers tend to be a certain sort of person, who also happens to like neighborhoods like mine? Are there seemingly lots of slender people in my neighborhood because they are influenced by other slender people, or are people predisposed to being slender moving to my neighborhood? The only way to ever know is to pick random people from across the country, who do not necessarily want to go to Brooklyn, and put them there.

  We shall return to this question in the next chapter, but for now let us consider something else that is happening at a site like Amazon as feedback accrues. Godes and Silva argue that as more reviews come in, “the more dissimilar a shopper is from the previous set of reviewers.” In other words, taste is happening. People are expressing their own disappointed reaction to a book, but they are also writing about other people’s enthusiastic reactions. As much as they are coming up against a thing they do not seem to like, the more unsettling occurrence is that they are coming up against people who do not seem to be like them.

  As the French sociologist Pierre Bourdieu argued, “Tastes (i.e., manifested preferences) are the practical affirmation of an inevitable difference. It is no accident that, when they have to be justified, they are asserted purely negatively, by the refusal of other tastes.”

  When the taste stakes are higher, the dissonance can seem even sharper. When a book, particularly a novel, wins a big prize, its reception by readers on the Amazon-owned user-generated review site Goodreads actually gets worse. Balázs Kovács and Amanda Sharkey, who performed the analysis, call it the “paradox of publicity.” It is not that judges are incapable of picking good books or that readers cannot recognize their merits. In fact, books that were only short-listed for an award had lower ratings than the award-winning books—before the prize was issued. Once the book was adorned with a sticker denoting the prize it had won, however, its ratings began to drop more precipitously than those for the short-listed books it had beat out.

  Why the backlash against this ostensible mark of quality? Prizes can increase book sales, but this is a double-edged sword. The prize, the authors note, raises expectations; it goes from being a book you might like to being a book you should like. The prize may also attract readers whose tastes are more mismatched than those of readers who weighed in before the book got the prize. This is the reverse of what had happened to Netflix, in the early twenty-first century, when its movie ratings went up across the board: The films did not get better; the algorithmic matching did. With Goodreads, when books “got better” because of having prizes bestowed on them, new, potentially less well-matched readers were drawn to them, like moths to a brightly burning flame.

  Not surprisingly, this often ends in dashed expectations. This might have been what happened with Groupon users on Yelp; they were more casual consumers drawn in for less than “natural” reasons. Unlike the more gently oscillating patterns of opinion the usual book on Amazon sees, the “shock” of the award triggered a spike in polarization: Not only did more one- and two-star reviews begin to enter the picture, but the number of “likes” on those reviews, as Kovács told me, “went up like crazy.” Middle-ground reviews—the landscape of meh—hardly moved the needle. Haters gonna hate, as it were. But haters also gonna rate.

  By analyzing some thirty thousand reviews, the study bore out a long-held truism: You can be well liked by critics or by the majority of readers but rarely by both. Longtime fans of the author in question may themselves balk at the newfound popularity bestowed on “their” previous favorite. The legendarily acerbic Chicago producer and punk musician Steve Albini once described the dynamic he was experiencing as Big Black, his small, obscure band, became more popular: “As the band gets bigger, you start having people show up to the shows who aren’t really of the same mind-set, they’re just there for a night out, you know?” They were people who were not only more indifferent to the band, he suggested, but indeed “people that would probably be hostile to you in a neutral setting.”

  There is a rather gloomy endgame looming here, though: the artist only prod
ucing art that people he likes will like, people only drawn to artists they think they will like. Does the world of online taste open us to new experience or simply channel us more efficiently into our little pods of predisposition?

  We are looking for “signals of trust” in the noise. When reviewers use their real names, their reviews are judged more helpful. What else drives that positive review of the review? As mentioned before, reviews that hew closer to the average number of stars are judged more helpful. Interestingly, that study found, the bias is not symmetrical: “Slightly negative reviews are punished more strongly, with respect to helpfulness evaluation, than slightly positive reviews.” When in doubt, we skew positive.

  But not always. There is one crucial variable that determines whether we like, or trust, negative reviews: whether something is an experience good (like a book or a movie) or a search good (a camera or replacement windshield wipers). While negative reviews in general were seen as less helpful than “moderate” reviews, as Susan Mudambi and David Schuff found in analyzing Amazon reviews, they were judged particularly harshly when the product was a book or a movie. Why? “Taste plays a large role in many experience goods,” they write, “and consumers are often highly confident about their own tastes and subjective evaluations, and skeptical about the extreme views of others.” Unlike with windshield wipers, people might have already made up their minds about a book or a film as they scan the reviews and can filter out someone’s one-star critique as a form of cognitive dissonance.

  In one of my favorite one-star reviews for Cormac McCarthy’s novel The Road, you can practically feel the reviewer trying to escape this trap with a defensive thrust:

  I know that many people love this book. Keep in mind that although you and I disagree, I am still providing information about the book that can be useful to people who have not yet decided whether to buy it. That is my purpose. I’m not trying to malign McCarthy or your personal taste, but to give a review from a different point of view.

  The reviewer is tap-dancing around taste, as if even mentioning it were to utter something indelicate. “Taste is a merciless betrayer of social and cultural attitudes,” observes Stephen Bayley, “more of a taboo subject than sex or money.”

  With search goods, people are looking for technical information, user tips, product flaws, and the like. They may have no biases or preferences, and a negative review may signal a tangible product flaw.

  The most extreme reviews for, say, an OXO salad spinner generally involve a product failure. But the one-star reviews for Rachel Kushner’s Flamethrowers, the (prizewinning) book I happened to be reading while working on this chapter, are filled with sentences like “I think my main issue is that I couldn’t relate to any of the characters.” Is that a flaw with the product or the reader? Books may fail or succeed, but not in the same way for every reader. To paraphrase Tolstoy, each unhappy reader is unhappy in his own way. On the other hand, people do not have trouble relating to their salad spinners. Then again, devices for drying lettuce are probably less personally reflective of people than the books they buy. As the business scholar Sheena Iyengar writes, “The less a choice serves some utilitarian function, the more it implies about identity.”

  Curiously, in the study of prizewinning books, the post-prize falloff in ratings was actually less for nonfiction books. Arguably, these are more utilitarian products, with fewer places for taste to intercede. It is as if we were almost instinctively wired to recognize the expression of other people’s taste when we see it—particularly when it diverges from our own.

  It makes one wonder whether all disliking is, however remotely, linked to the primal disgust mechanisms mentioned earlier with food. Indeed, when people in one study looked at negative reviews of different products—“utilitarian” and “hedonic” goods—they were more likely to attribute the reasons for negative reviews to something about the thing when the product was utilitarian and to something about the person when it was hedonic. “Taste classifies,” wrote Bourdieu, “and it classifies the classifier.” And then we classify the classification.

  —

  You might argue that reviewing restaurants at Yelp and books and salad spinners on Amazon and films on Netflix are all different things. And yet a curious meta-logic takes over online. People are generally not situating a work in its historical context or doing the other kinds of heavy lifting that critics were once paid to do but reflecting upon their own consumption experience.

  A “content analysis” of movie reviews on one Web site, looking at differences between critics and online “word of mouth,” found that critics talked more regularly about plot, direction, and acting than average moviegoers (when they refer to themselves, interestingly, their reviews deviate more from the average; nothing signals taste more than the word “I”). Amateur critics, meanwhile, talked more about the personal relevance of a movie (in 33 percent of films, versus zero for critics). In nearly half of moviegoer reviews, the reviews responded to film critics (quite naturally, no critics talked about average moviegoer reviews).

  In short, critics talk about why something should (or should not) be liked; people talk about why they did like something. Curiously, critics are often criticized for trying to impose their own taste on the “rest of us,” when actually it is the “rest of us” who are most guilty of this practice.

  People are now so accustomed to the reviewing mind-set that one occasionally spies a flummoxed “review” of a simple product like paper clips: “What can I say? They’re paperclips!” Four stars! That a site like Amazon sells virtually everything tends to blur and flatten things. Books are savaged because they are not available as an e-book or for their typeface. The lines of authority are made muzzy. What does the competent paper clip critic have to say about French Symbolist poetry? What are the criteria for reviewing something such as a noise machine, and from where does authority come? (One actual line: “The white noise is a little too deep for us.”) The rise of online reviewing may be toppling the singular critical voice from its pedestal. But with its fall, taste has shattered into a thousand fragments. We are sifting through those shards, trying to make meaning of other people’s attempts to say what something meant to them.

  Next, we shall flip the question over: not what you say about your choices, but what your choices say about you.

  * * *

  *1 O’Brien’s volume itself has a 3.75 out of 5 stars ranking on Goodreads.​com.

  *2 Time has been kinder to the movie. On IMDb.​com, it has a 6.9 out of 10 rating.

  *3 Language inflation is another problem with online reviews. The creators of RevMiner, an information extraction app designed to streamline Yelp, note that a person searching for something like “good dim sum” does not really mean good dim sum, but “dim sum that others have described as ‘great’ or ‘amazing.’ ” Good is no longer good enough. You need to be awesome.

  CHAPTER 3

  HOW PREDICTABLE IS OUR TASTE?

  WHAT YOUR PLAYLIST SAYS ABOUT YOU (AND WHAT YOU SAY ABOUT YOUR PLAYLIST)

  He guessed at intense little preferences and sharp little exclusions, a deep suspicion of the vulgar and a personal view of the right.

  —Henry James, The Ambassadors

  LOST IN TASTE SPACE

  Who does Google think you are?

  There is an easy way to find out. Type “http://​www.​google.​com/​ads/​preferences.”

  The search company believes that I am an English-speaking male, age twenty-five to thirty-four, with leading interests in “air travel” and “Action & Adventure films.” “Well, now,” I think, “how useful could this be? It thinks I’m more than ten years younger than I really am!” But then a darker realization sets in: It could be that I am simply acting ten years younger than my age. All my Google searches have boiled me down to a person who flies a lot and watches action flicks (often at the same time). “You don’t know me,” I want to protest, with an air of Ray Charles anguish, but perhaps I do not know myself as well as my
idealized self. Having this portrait played back at you can be as unsettling as seeing your reflection in the screen of your smartphone; is that really who I am?

  We are, of course, more than our search terms. How much can be inferred about me by my search for printer toner replacement cartridges, other than that I am a person in need of a new printer cartridge?

  As Hugo Liu, chief data scientist of the recommendation start-up Hunch.​com, told me one afternoon over coffee in New York City’s Chelsea neighborhood, “If someone happens to search for cats a lot on the Internet, or if someone’s looking for a stroller part, how much of that is taste?” Liu, who, with thick black glasses and an artful pile of tousled hair, affects a mad data scientist look, has long puzzled over the question of how to extract, model, and predict patterns of people’s behavior online. As a student of the MIT Media Lab’s Pattie Maes—who, among many other things, developed the pioneering Firefly collaborative filtering recommendation systems—he was bothered by those systems’ lack of dimensionality. “They hint at people but don’t really represent them in any way. It’s my behavior in a particular domain,” he tells me. What I bought at Amazon, what I watched on Netflix. “But what if I could create a model of people that could work across domains?”

  In other words, what if you could meld what you watched on Netflix with what you listened to on Pandora; marry that up with what you bought on Amazon and other online retailers; then overlay that with the people who interested you on Match.​com and the food you bought last month; then mix in myriad personal details—the way you talked, the images you saw on a Rorschach test, your beliefs in science and God—and then take all that and correlate it with data from millions of other people. Might you just then begin to have a more robust way of understanding people as a tangible variable? At the heart of the question lay a larger one: Just how predictable is our taste?