“The Bestseller Code,” a new book in which Jodie Archer and Matthew L. Jockers present an algorithm for detecting the sales potential of other books, has, not surprisingly, a commercially canny title. “The Da Vinci Code,” after all, has sold more than eighty million copies. Of course, the success of that novel may have had less to do with the title than with the conceptual hook (murder, sex, conspiracy, and Jesus) or Dan Brown’s bottomless bag of chapter-punctuating twists. But in “The Bestseller Code” all correlation is causation, and titles matter. (So do, for instance, semicolons.) Titles pointing “to things, to objects, typically common nouns,” appear often on best-seller lists, Archer and Jockers explain. Sometimes the nouns come with a qualifying word, like “Bestseller” or “Da Vinci,” but the most best-seller-y option lets the nouns stand alone. Archer and Jockers point to “The Goldfinch” and “The Firm” as particularly primo titles, making you wonder why they didn’t call their own book “The Code.”
But then most books aren’t written to please a success-predicting algorithm—not even this one. Archer and Jockers focus their study on fiction, and the unfriendly economics of fiction-writing generally insure that a deep, irrational passion drives most of its authors. There is no possible insight into the common denominators of best-sellerdom that could turn writing a first novel into a judicious economic decision; there’s no algorithm that would change the fact that most Americans read just a handful of books a year, and a quarter of them don’t really read books at all. In any case, “The Bestseller Code” doesn’t pretend to hold the keys to the top spot: it just claims that the books that sell millions tend to be predictable, and not in the ways that you think.
Archer and Jockers, who met at Stanford and come from a publishing and an academic background, respectively, built and refined their algorithm over the course of four years. By the end of that time, they claim, it could deduce with eighty per cent accuracy whether or not an unmarked manuscript had hit the New York Times best-seller list. It assessed novels by Jessica Knoll, Mitch Albom, Chad Harbach, and Michael Connelly and identified their best-selling chances at upward of ninety per cent. Archer and Jockers argue that this ability to predict mainstream appeal across a heterogeneous sample makes the algorithm a clearer reader than many a critic and acquisitions editor—and that it has detected a “distinct set of subtle signals” separating best-sellers from the rest of the bunch. The authors acknowledge, but do not give much credit to, the influence of reviews, splashy covers, big-name blurbs, and marketing budgets; in their scheme of success, postproduction doesn’t matter—the things that really make a book a best-seller are already embedded within the text.
That logic seems narrow, but not tremendously controversial—and, while it’s interesting to see the data of aggregated attraction, the conclusions that can be drawn from the data are much less subtle than Archer and Jockers seem to think. Readers, we learn, want a colloquial style, a decisive main character, a fast-moving, rhythmic plot. In a section on topic and theme, Archer and Jockers explain that people like to read books with a small set of central topics, and that it’s good if those topics are familiar and if they contrast with each other in an interesting way—crime and domesticity in “Gone Girl,” for instance. Later in that section, after half a page of dramatic buildup, they identify the topic most predictive of a best-seller: “human closeness and human connection.”
At times, it seems like Archer and Jockers are trying to retrofit a closed system. They found that best-sellers have lots of contractions—the better, they explain, to mimic contemporary speech—and exclamation points only rarely. “Top-selling authors know that there is nothing more annoying than something like ‘It was getting dark! The stairs creaked! Maybe there was a ghost!’ ” they write. They conclude that best-sellers consist of “shorter, cleaner sentences, without unneeded words,” and that best-selling characters “make things happen.” Active verbs predict best-sellers better than passive verbs. “Hesitation doesn’t keep pages turning,” Archer and Jockers decide. After all that work, in other words, the algorithm ends up confirming the uncontested tenets of craft and style.
Nonetheless, there’s an awkward charm in watching an algorithm discern the things that humans appreciate instinctively. In a section about syntax, Archer and Jockers point to “Reader, I married him,” Charlotte Brontë’s famous line. “Isn’t the entire point of so many stories to get that ‘I’ and that ‘him’ closely aligned, separated by an all-important verb like ‘married’?” they write. “So often, this is entirely why we keep turning the pages.”
A best-seller code, packaged and licensed to wannabe authors and ambitious/avaricious/or some such publishing houses, could be very useful. I, for one, would rather pay a program to analyze my plotting rhythm than sit through the average creative-writing workshop. It’s appealing to imagine that Archer and Jockers have built an intelligent catalogue of the American reader’s taste and will—a machine reader that would never have an off day, or take a pass on the next John Grisham, or care too much about a fetching head shot, or discriminate by age or race or weight. The algorithm, Archer and Jockers write, could keep the publishing industry “not just running but diverse.”
Similar attempts to introduce algorithmic judgment have been made in contemporary pop music, a genre where best-selling stylometrics are clearer, are mathematically transcribable, and have been central for some time. There’s a nearly fixed structure in pop music, and a best-selling chord progression, I-V-vi-IV. (That’s the cathartic loop you find in Toto’s “Africa,” Miley Cyrus’s “Party in the USA,” and the Beatles’ “Let It Be.”) Top Forty radio tends to average out at around a hundred and twenty beats per minute; hit songs last between three and four minutes; the list goes on.
Because math in music is clearer, attempts to capitalize on it have moved further along. In 2009, a company called Music Intelligence Solutions released commercial software, called uPlaya, that allowed musicians to upload a track and receive, via proprietary “Hit Song Science,” a catchiness score. Here, one sees the limits to algorithmic judgment in the fact that, for example, uPlaya gave an 8.9 (on a scale of 10) to the commercially successful and powerfully irksome “I Got a Feeling,” by the Black Eyed Peas. That track’s chord progression would test well; a program could match it up with Fall Out Boy’s “Sugar We’re Going Down” and the verses of Taylor Swift’s “Fifteen.” It clocks in at a hundred and twenty-eight beats per minute and generally hews to the math. But there’s nothing communicated by that 8.9 that a person with the most rudimentary musical knowledge can’t hear immediately; what’s more, only a human listener would hear that “I Got a Feeling,” with every replay, grates increasingly on the ears. When the data are reliable, an algorithm may be able to predict success, but it can’t tell you what’s good. In any case, nothing of note happened with uPlaya: the last post on its Facebook page reads, “It looks like the music industry is not as bleak as predicted,” and is dated January 30, 2012.
At the beginning of “The Bestseller Code,” Archer and Jockers concede the algorithm’s limits. “Our belief,” they write, “while it may be irritating and old-fashioned, is still that if you want to be a bestselling writer then first you have to learn and really appreciate fiction with as many tools as you can.” That belief may be old-fashioned, but it’s not irritating. It does, though, threaten the utility of a code: once you’ve done that learning, you don’t need an algorithm to confirm it.
