Amazon's book recommendation engine is dumb -- any better ones?

Cafe Society
1

Like a lot of Dopers, I’m someone who reads voraciously but selectively: I always need to be in the middle of at least one novel, I typically finish at least two or three novels a week, and while I enjoy a broad range of genres/authors/styles, I have a pretty definite sense of what I like, and I have no qualms about putting down a novel at any point – beginning, middle, or end – if I’m not finding it entertaining or absorbing.

Naturally enough, this means that I’m always on the lookout for new books to read. When I can’t get to the public library or browse used book stores, I trawl Amazon for ideas and recommendations by free-associating from books I like and seeing what catches my eye (Amazon Prime is a wonderful thing, and I’m pretty sure I’ve spent way too much money on that site as a result).

The problem, though, is that Amazon’s recommendation engine – and its related “Customers Who Bought This Item Also Bought” feature – really, really suck. With few exceptions, the recommendations it comes up with invariably fall into four less-than-helpful categories:

[list=A]
[li]other books by the same author, which – besides being pretty useless as recommendations to anyone with an IQ over 4 – take up valuable space when the book you’re using as your touchstone is by someone like Terry Pratchett or Diana Wynne Jones who have an oeuvre of two dozen or more[/li][li]the same couple handfuls of most-bought books in a particular genre, repeated again and again for whatever genre you happen to be looking for recommendations in (The Accidental Time Machine, The Graveyard Book, Halting State, and Anathem, among others, show up as recommendations for every single science fiction book you can think of, for example)[/li][li]the same handful of most-bought books from other genres, repeated again and again irrespective of the genre you’re searching in, since the algorithm that generates the recommendation lists doesn’t take into account the fact that a bestseller like The Guernsey Literary and Potato Peel Pie Society or The Girl With the Dragon Tattoo (both of which appeared among the recommendations for Bellwether by Connie Willis) or The Brief Wondrous Life of Oscar Wao (recommended if you like Perdido Street Station by China Mieville, naturally) will, by virtue of their best-sellingness, appear as something that “Customers Who Bought This Item Also Bought” even if it has nothing whatsoever in common with the “This Item” in question[/li][li]multiple books from one or two series by one or two other authors in a particular genre, repeated again and again for whatever genre you happen to be looking for recommendations in (although with a little more variability than the most-bought books of that genre[/li][/list]

All of which means that even though Amazon carries hundreds of thousands of different titles, someone looking for recommendations is only likely to see two or three hundred of those within a particular genre or cluster or authors…and they’ll see that same two or three hundred again and again and again.

There must be tons of books out there for sale that I’ve never heard of, but that are exactly the kind of thing I’d be likely to enjoy. Why doesn’t Amazon make it easier for me to spend my money on them?

…If anyone has any suggestions for a book recommendation engine that doesn’t just give me the same old stuff, I’d be interested to check it out.

2

This one clusters by user input, so it would likely suit your needs better.

Gnooks.

3

Posting to subscribe. I have the same complaint. If I like a particular author I have likely already checked out his other books - don’t bother me with those recommendations. Likewise, if I’ve said no to a particular author several times, stop recommending him!

4

I’d pay extra if they would allow us to once-and-for-all close off some avenues of recommendations. I’m never going to buy a computer game. Stop suggesting them. There are some authors I refuse to read anymore. Stop suggesting them. I’m not going to buy the calendar, t-shirt, coffee mug or trading card of whatever. STOP showing them to me!

5

This interested me too, so I looked for alternatives that I might try in the future, since gnooks doesn’t work with titles.

Two promising candidates are What Should I read Next? and Goodreads.

6

Ignore those, but take a look at the “Listmania!” and “So You’d Like To…” links that often appear near the bottom of the page for a book you like. That way you can see specific recommendations made by other customers who liked that book.

7

I read and rate books that are written for children/teens/young adults. I like some Disney movies, and have rated them.

I do not want anything that is marketed as part of the Disney Princess line. Nor do I want to have anything to do with Barbie.

Similarly, while I read and enjoy a lot of SF/fantasy, I absolutely will NOT buy anything by Robert Jordan or Anne Rice or a whole host of other writers. I read a lot, but I’m damn picky. If I was locked in a cell, and had an Anne Rice book to read and nothing else, I’d contemplate the lint in my bellybutton.

8

The thing that kills me about Amazon’s book recommendation isn’t any of that. My problem is that Amazon constantly recommends books that I already own. That wouldn’t be so bad except that I own the books because I ordered them from Amazon.

9

I’ve never seen any of those books show up on my recommendations list, and I’ve bought many SF books from Amazon, and told it I own many more.

Why is this a problem? If a book is a best-seller, then that means that a lot of people like it. You might be one of those people. I’d be frankly puzzled if a book-recommender didn’t include a fair number of bestsellers.

The only real problem I’ve found with Amazon’s recommendations is that it considers every edition to be a completely different book. If I’ve already told it that I own Fellowship of the Ring, The Two Towers, and Return of the King separately, then I’m not going to be interested in Fellowship of the Ring: Movie Art Cover Edition, or Lord of the Rings Boxed Set.

10

Chronos:

As soon as I finish this assignment (next hour or so), I’ll put up a detailed post comparing recommendations and explaining exactly what I mean.

11

Okay. Here’s an in-depth look at Amazon’s recommendations for two books: Bellwether by Connie Willis and Perdido Street Station by China Mieville. I chose these two because although they both fit vaguely beneath the umbrella genre of SF/Fantasy (and although they’re both excellently written and extremely enjoyable, in my opinion), it’s hard to imagine two novels that are more different.

Bellwether, for example, is a witty, frothy, screwball comedy set in Colorado in the late 1990s (the book was written in 1997). Although its main characters are scientists and it largely takes place in a research lab, it barely has any SF or fantasy elements at all (I can think of two aspects of the book that are even remotely fantastical, and both are tangential plot devices that are first introduced in the book’s denouement). I’m pretty sure it’s only considered SF/Fantasy by virtue of Willis’s other books. Oh, and it weighs in at a paltry 247 pages.

Here’s an excerpt from one of Amazon’s blurbs:

By contrast, Perdido Street Station takes place in a phantasmagorical steampunk-y cityscape on some distant planet. It deals heavily in urban fantasy (among other things, there’s magic and druglords and political dissidents and oppressed minorities and mutilation and weird scientific contraptions and inter-species sex), and although there are certainly amusing passages here and there, no one would ever mistake it for a comedy. If anything, it’s a social discourse on the human condition (as well as an incredibly detailed feat of world-building). At 640 densely written pages, it’s almost three times as long as Bellwether, and feels longer.

Here’s an illustrative blurb (no real spoilers):

In my next post I’ll take a look at the recommendations Amazon gives for Bellwether and compare them to the recommendations given for Perdido Street Station.

12

Goodreads.com doesn’t have a recommendation engine, but if you build up a list of Friends with similar tastes you can track what they are reading and compare your library to theirs. There are also a lot of Groups, based on genres or types of books, and that is a good place to get recommendations. (There are some Dopers there, to be found in this group.)

13

Let me say at the outset that I understand that the “Customers Who Bought This Item Also Bought” feature does not claim to give recommendations for other books you might like. What it does do, naturally enough, is list one hundred or so other things that were purchased by the people who purchased the thing you’re looking at. However, there’s no reason to provide a feature like this unless it was intended to furnish recommendations of other things to buy, especially since there’s really no other way to plug in the name of a book on Amazon and find other books that might be similar. (The Listmania feature is extremely erratic and generally unhelpful.) So I’m going to use the “Customers Who Bought This Item Also Bought” feature as a proxy for Amazon’s book recommendation engine for the purposes of this post, although I’m willing to listen to arguments that I’m trying to measure the wrong thing or that I’m taking Amazon’s engine to task for something it wasn’t designed to provide.

In any event, using the “Also Bought” feature for Bellwether yields 99 results. Of those 99 results, 14 are for other books by Connie Willis. I think we can all agree that as recommendations go, those are pretty much useless.

(I should add at this point that I’m going to be eyeballing the books on this list to see whether they seem like plausible recommendations. There’s no particular criteria I’ll use to gauge that, but I don’t expect my conclusions to be very controversial one way or the other. If there’s one overarching aspect of Bellwether that similar books might be expected to share, it’s the screwball wittiness. Bellwether is intended, first and foremost, as a comedy.)

The remaining 85 results break down as follows:

Two books by Ilona Andrews about “a mercenary who cleans up after magic gone wrong” in an alternate-universe Atlanta. Beyond a vaguely contemporary setting and a female protagonist, these books seem to have absolutely nothing in common with Bellwether in tone, style, subject matter, theme, or substance.

The Handmaid’s Tale by Margaret Atwood. Um…near-future dystopia involving a tyrannical theocracy and an oppressive gender caste system. I guess Margaret Atwood and Connie Willis are both women; maybe that explains it. (Actually, I’m guessing that this is on the list because it’s a Genre Classic. The customer base for a Genre Classic is broad enough that it will be bought by a ton of people who also buy Bellwether even if the Genre Classic and Bellwether have nothing whatsoever in common.)

In the Garden of Iden by Kage Baker. Time traveling cyborg botanists, identity crises, and tea parties. Not particularly similar thematically or stylistically, but it’s witty enough, and I can at least see where they’re coming from.

Four books by Iain M. Banks (Consider Phlebas, Excession, The Player of Games, and Use of Weapons). Installments of the Culture series, set in a sprawling galactic empire of the far future. Great books with more whimsy than you might expect, but not particularly similar to Bellwether at all.

Three books by Patricia Briggs about a “kick-ass supernatural heroine.” Basically the Ilona Andrews books above, but with werewolves.

People of the Book by Geraldine Brooks is about a rare-book expert (who’s a woman!) restoring an ancient Jewish manuscript in 1996 Sarajevo. Looks like literary fiction, and a Bestseller. Bestsellers, like Genre Classics, are bought by enough people that they’ll show up on these sorts of lists even if they don’t have anything in common with the book in question. (And regardless of the merits of the Bestseller itself, if it doesn’t have anything in common with the book in question, then the recommendation is useless to someone looking for books that are similar to the book in question.)

World War Z by Max Brooks. Another Bestseller.

Orphanage by Robert Buettner. Military science fiction/space opera. Nothing in common with Bellwether except that they’re both bound volumes of text.

Three books by Lois McMaster Bujold. Here’s the description given for one of these books: “In this two-part story, Cordelia Naismith, made an outcast after being forced into marriage with her arch enemy, finds further trouble when her husband is made the guardian of the infant heir to the imperial throne.” Yup. (Again, not saying that these are bad books or that Bujold is a bad author, only that as book recommendations for people who liked Bellwether, they’re not particularly on point.)

Blood Rites by Jim Butcher. One of The Dresden Files books, which I’ve heard good things about. I’d be surprised if it’s something you’d expect to appear on a recommendation list for Bellwether, though.

Five books by Jack Campbell. The Lost Fleet series – more military science fiction with spaceships and shooting and stuff.

Yiddish Policemen’s Union by Michael Chabon. Another Bestseller.

Little Brother by Cory Doctorow. Fun book. Good sense of humor. Similar to Bellwether in that neither is particularly grounded in SF or fantasy. This is not a terrible recommendation.

Orbiter by Warren Ellis. Graphic novel about a space shuttle that returns to Earth rewired with alien technology and the people who try to figure out what happened.

Two books by Jasper Fforde. These are from the Thursday Next series. Frothy, witty…good recommendations.

Eifelheim by Michael Flynn. Present-day scientist couple investigates the crash of an interstellar ship in the 14th century. The book is described as a “heartbreaking morality play of stranded aliens in medieval Germany,” and it’s #16 on Amazon’s sales list in the science fiction category, so I’m tempted to call this a Genre Bestseller. The plot conceit has enough similarity to Bellwether, though, that I’m willing to give the benefit of the doubt and call this a decent recommendation.

In the Woods by Tana French. Great book, but it’s a Scottish police procedural/thriller, and I have no idea what it’s doing on this list.

Five books by Neil Gaiman, including The Graveyard Book. Eh…Gaiman is witty enough that I actually think his books are decent recommendations for people who like Connie Willis. Still, I’m pretty sure that you’re going to find a lot of Gaiman on any suggestion list for a book with some element of urban fantasy or humorous SF/fantasy.

Three books by Joe Haldeman. One of them is The Accidental Time Machine, which is a Genre Bestseller, another is The Forever War, which is a Genre Classic, and the third is military science fiction.

Three Men in a Boat by Jerome K. Jerome. A genuinely excellent recommendation, although its appearance on this list can mostly be attributed to the fact that a different Connie Willis book directly references Three Men in a Boat throughout. It was nice of Amazon to put three different editions of this book on the Bellwether suggestion list, though.

Little Heathens by Mildred Kalish. Nonfiction memoir about a childhood spent in Iowa during the Great Depression. Not sure why it’s on here.

The Girl with the Dragon Tattoo by Stieg Larsson. Bestseller.

The Left Hand of Darkness by Ursula K. Le Guin. Genre Classic.

The Lies of Locke Lamora by Scott Lynch. Genre Bestseller, although it’s heck of a lot of fun and wouldn’t be a terrible recommendation, I guess.

The Engines of God by Jack McDevitt – Hard SF involving alien ruins on a planet called Quraqua.

Canticle for Liebowitz by Walter Miller. Genre Classic.

We3 by Grant Morrison. Graphic novel about bioengineered animals.

The Time Traveler’s Wife by Audrey Niffenegger. Wonderful book, but not sure what it has in common with Bellwether. Bestseller?

Fleet of Worlds by Larry Niven. Hard SF set in the year 2650.

Gateway by Fredrik Pohl. Genre Classic, I’m guessing.

Nine books by Terry Pratchett. See my comment on Neil Gaiman above.

The Name of the Wind by Patrick Rothfuss. Genre Bestseller.

Two books by Maria Doria Russell. They apparently tell the story of the story of “Father Emilio Sandoz, the Jesuit priest whose faith was brutally tested . . . on the faraway planet of Rakhat.”

The Hero of Ages by Brandon Sanderson. Epic fantasy.

Three books by John Scalzi. Military science fiction, and at least one of them could probably be considered a Genre Bestseller as well.

The Guernsey Literary and Potato Peel Pie Society by Mary Ann Shaffer. Bestseller.

Three books by Neal Stephenson. I’d say two are Genre Classics and one is a Genre Bestseller.

Earth Abides by George R. Stewart. Post-apocalyptic saga; Genre Classic.

Dies the Fire by S.M. Stirling. More post-apocalyptic stuff. Genre Bestseller?

Creating a Newsletter in InDesign by Katrin Straub. Um…

Two books by Charles Stross, one of which (Halting State) is a Genre Bestseller. (The other one is off-beat science fiction involving armies of zombies and British bureaucracy, which makes it a far better Bellwether recommendation than most of the titles on this list.)

Four books by Vernon Vinge. Far-future hard SF.

Blindsight by Peter Watts. Hard SF.

Two books by Robert Charles Wilson. Wilson generally takes present-day Earth, introduces one seemingly inexplicable SF element, shows the consequences to everyone’s mundane lives, and lets his protagonists (who are often scientists) try to figure it all out. That’s close enough to Bellwether that I’ll call these decent recommendations.

The Book Thief by Markus Zusak. Bestseller.

And finally, the Dr. Horrible’s Singalong Blog DVD, which is a Genre Bestseller but has enough of Bellwether’s sensibility that I’ll call it a recommendation.

Summary: Of the forty-five different authors represented above, only ten of them at most have books listed that could plausibly be considered recommendations for someone who likes Bellwether. The rest of the list consists of a ton of random military science fiction, paranormal ass-kicking, far-future epics, bestsellers, and genre classics that have no more in common with Bellwether than they would with any other randomly selected book in the SF/Fantasy section of Amazon.

I think that’s a pretty poor showing by the recommendation engine, frankly, and I haven’t even gotten around to comparing the books on this list with the books on the Perdido Street Station list. (Hint: only twenty-four of the authors on the Bellwether list don’t also show up on the Perdido Street Station list, and that includes the woman who wants to teach us all about InDesign.)

14

I like Bellwether, and I’ve read quite a few on that list.

It looks to me like they’re giving you recommendations based on the author rather than the particular book. Connie Willis links to sci-fi written by women, screwball comedies and time travel. That explains a lot of the entries. She has also written some post-apocalyptic short stories, so that may explain a few more.

The urban fantasy (Butcher and Briggs are both pretty good, by the way) doesn’t seem to belong. Except maybe they figure that women who read science fiction might be willing to dip into it.

15

What metric could one possibly use that would be better than “Customers who bought this also bought…”? If a lot of people who buy Bellwether* also buy paranormal ass-kicking books, even if there’s no obvious reason why the same people bought both, then that’s a good recommendation.

16

Actually, I think they’re doing exactly what they say: giving you a list of things purchased by people who purchased Bellwether from Amazon. It’s perfectly possible that people who purchase Bellwether gravitate towards sci-fi written by women, hard military science fiction, and the rest of it. I’m at a loss to explain the InDesign manual, though – that can’t possibly be among the top 100 items purchased by Bellwether purchasers.

17

Despite the trendiness of werewolves and vampires these days, I urge you to read these books. Briggs is quite a good writer. I think that she’s consistently better than Willis, even. I loved some of Willis’ books, but at least a couple of them have left me full of meh. Lincoln’s Dream and Passage, if you must know.

18

You really don’t understand the Venn diagram aspect of this?

Let’s take a hypothetical bestseller called Slice of Pi. Say that forty percent of the reading public owns a copy of Slice of Pi, but only twenty-five percent of the people who own a copy of Bellwether also own a copy of Slice of Pi. (Let’s also say, for simplicity’s sake, that if you own a copy of a book, that means that you like it.) This strongly suggests that people who own a copy of Bellwether are significantly less likely to like Slice of Pi than people who don’t own a copy of Bellwether. Yet as long as that 25 percent is still a large enough chunk of the Bellwether-owning public, Slice of Pi will show up on the list of items that people who purchased Bellwether also purchased…even if the other 75 percent of the people who own Bellwether would think that Slice of Pi is bland, allegorical crap.

Your conclusion that Book A will be a good recommendation for someone who likes Book B as long as x percent of people who own Book B also own Book A just isn’t logically sound.

Lynn: I’ll have to check the Briggs books out. I liked Passage a lot, but not so much Lincoln’s Dreams.

19

And when I get home tonight, Chronos, I’ll post the results of the Perdido Street Station list – a third of which is identical to the Bellwether list, and another twenty percent of which consists of other books by authors who are on the Bellwether list.

20

By the way, Connie Willis is working on a new novel in her time-travel universe called All Clear, about the London Blitz. She is supposedly in the editing phase.