icon caret-left icon caret-right instagram pinterest linkedin facebook twitter goodreads question-circle facebook circle twitter circle linkedin circle instagram circle goodreads circle pinterest circle

From Phyllida's Desk

Algorithms and Genres

People make fun of the Netflix algorithm, the computer program that recommends movies based on how you rate other movies. I’ve certainly laughed at it. Its worst recommendation for me was Jackass 2. And no, I didn’t give Jackass 1 five stars, or even watch it.

The reason Netflix suggested Jackass 2 was that I had given Monty Python’s Life of Brian five stars, as did, apparently, many jackasses.

Netflix’s algorithm tries hard, and its rating system isn’t exactly complicated. Three stars means you “liked” a movie, four that you “really liked” it and five that you “loved” it. Two stars is “didn’t like” and one star is “really didn’t like.” Netflix prompts you to rate every movie you rent or stream, and when it recommends a movie, it always gives you a reason or two: “Because you enjoyed—“.

But there’s no way for viewers to indicate why they liked a particular film, to distinguish among all the variables that go into making a good movie: story, actors, dialog, cinematography, direction. Which is, of course, the strength of the system. All that matters to Netflix is finding other movies that you will “really like.”

Walter Kirn, in Sunday’s “The Way We Live Now” column (New York Times magazine), talks about how Netflix has created customized headings for his household’s preferences, such as “Cerebral Romantic Foreign Movies” (or as he calls them, “pretentious sex films”) and “Goofy Suspenseful Action and Adventure” that his kids watch—and then talks about how, in fact, a goofy suspense film might suit him better than some of the more “cerebral” choices Netflix made for him (e.g., Radio Days).

I’ve long had fantasies of messing with the Netflix algorithm’s head (if it has a head), because I “really like” so many seemingly disparate kinds of movies. But all that I’ve accomplished is to give Netflix a wider range of suggestions—which is the whole point of the exercise. Yes, I like “Gay and Lesbian Comedies” and “Witty Romantic Independent Movies.” My taste in Drama is so broad that at the moment Netflix has given up any attempt to subdivide. But at least I have a checklist of drama subcategories to choose from, including “Gritty,” “Understated,” and yes, “Cerebral.” Whatever kind of movie I’m in the mood for, Netflix will make an attempt to satisfy me, to find something that I’ll want because last night, or last week, or six months ago, I gave similar films four or five stars.

The main thing about Netflix is: it always tries. By contrast, my Kindle only makes an effort for genre fiction. I’ve had my trusty old first-generation Kindle for quite some time now, and I’ve built up a good-sized library. But when I go to shop in the “Kindle store,” only the genre titles generate new suggestions. Romance novels, whether gay, m/m, or hetero; thrillers and mysteries; and some historical fiction have prompted Kindle to suggest more of the same. But “literary” titles—Ann Patchett’s Bel Canto, Kazuo Ishiguro’s Never Let Me Go, Aravind Adiga’s The White Tiger—none of these budged the algorithm’s needle a millimeter away from romance authors Nora Roberts and Jo Beverley; Kathy Reichs (of the “Bones” novels and the TV series based on them); and the selection of gay and m/m titles that have been there for weeks. After each new literary title purchased, the 24 recommendations stayed exactly the same.

As my reading tastes have shifted, it’s as if the Kindle can’t get out of some snowdrift, still spinning its wheels in last week’s frozen slush.

The reason, of course, is that old theme of mine, classification. Readers who’ve been following me since Phyllida and the Brotherhood of Philander came out, back in 2008, have been down this road before, but it’s interesting to think about the relationship between genres and algorithms. The Library of Congress assigns subject headings to fiction, but public libraries and bookstores don’t arrange the “fiction and literature” section on the shelves by these subjects. Writers of literary fiction rarely confine themselves to one narrow subject over the course of a career, nor do most readers of literary fiction choose their books purely by subject. If I enjoyed, as I did, the Patchett and Ishiguro titles mentioned above, I’m more likely to look for another book by Patchett or Ishiguro, rather than another book about unusual hostage situations in foreign countries or boarding schools for cloned organ donors.

But libraries and bookstores, like Amazon, do use broader categories—genres—to separate out romances, sci-fi and fantasy, and mysteries. And search algorithms like Amazon’s and Netflix’s use genres, not narrow subject headings.

An LC subject heading is meant to be as specific as possible, essential in the world of nonfiction. In the AMNH Library, there are literally thousands of titles on “Paleontology,” 124 under “Paleontology—Mesozoic,” eight under “Paleontology—Mesozoic—China,” and three under “Paleontology—Mesozoic—China—Liaoning Sheng.” A scientist looking for a book on the feathered dinosaurs recently discovered in Liaoning Province, perhaps the earliest direct ancestors of birds, doesn’t want to plow through all 3300+ books on paleontology in general; or the 554 books on dinosaurs. She’s needs those precise LC subject headings to locate the three relevant titles.

Reading fiction for pleasure is different. We want, if not thousands of suggestions, more than two or three. Sometimes we find a writer we adore and we gobble up his or her works greedily, like a food supplying a nutrient missing from our bland, starchy diet. If we like genre, we can identify other promising titles and authors based on what we’ve read. Amazon can easily find more gay romances, historical mysteries set in Victorian London or ancient Rome, or forensic crime procedurals, if that’s my preference.

With literary fiction, we’re more often looking for a mood, a voice, a style, rather than any particular kind of plot or story. After an especially absorbing book, the last thing we want is another one “just like it.” Chances are we’ll want something a little different. We need time to savor what we just finished; if the last book was particularly harsh or intense, we may need a period of mental healing. But that doesn’t mean we don’t want to read—to escape—at all.

Some readers, like some writers, really do want to read or write the same thing over and over. The rest of us are constantly casting about for recommendations, and there’s no algorithm to help us. We read blogs and reviews, and rely on word of mouth and serendipity. I discovered the fabulous Sarah Caudwell’s four (sigh!) deliciously stylized, witty mysteries through a fellow member of the Jane Austen Society. I found my last truly fantastic read, Emma Donoghue’s Room, through that most old-fashioned of methods, print reviews. But sometimes nothing clicks, nothing tickles our fancy.

I got my latest good read from a torn-out magazine page that my boss left on my desk last week. I don’t even know which magazine. She gave it to me because we had been talking about The Phantom Tollbooth, and this page had a short article about how author Norton Juster and illustrator Jules Feiffer were collaborating again, fifty years later.

When, desperate for the printed word on my lunch hour, I turned the page over, there, under “Arts, Review of reviews: Books,” was a Novel of the Week: I’d Know You Anywhere, by Laura Lippman. The mini-review mentioned an earlier novel, What the Dead Know. And that’s what I read, on my Kindle, starting last night and finishing today, because, as the quoted reviewers said, Lippman is a writer who not only “understands the human heart,” but writes “hypnotic” thrillers that “equal the best of today’s literary fiction.”

Now if only Kindle could create an algorithm for that.
Be the first to comment