Disclaimer: This post comes from me as a technologist and a reader, not as a writer. As a writer, I keep my opinions on reviews and reviewers to myself.
All the recent exposure on GettingBookReviews.com and John Locke’s lack of disclosure on using the service prompted me to finally put up a post on something I’ve been thinking about for a while.
Paying someone to review your book is like asking your mom what she thought of the card you made her for Mother’s Day in the third grade. Wait. No. It’s more like going to The Moonlight Bunny Ranch, getting the all-you-can-@#$% special, and asking each of your escorts if you’re well endowed.
Don’t expect any honesty.
It’s clear we the readers have a problem. You see, people are also doing this without money changing hands. With the overwhelming number of people getting into self publishing, each trying to clamour for attention and sales, we’re also seeing a rise in highly-biased reviews from a variety of sources. The bottom line is you just can’t trust reviews the way originally intended. Now I’m not saying all self-published authors engage in this behavior, or that this behavior is isolated to only self-publishing, but the review landscape is tainted such that it’s often difficult to distinguish honesty from marketing.
Even Amazon’s own recommendations are questionable. Check out what this 2010 post from The Boston Review had to say (emphasis mine):
Jeffrey Lependorf, Executive Director of the Council of Literary Magazines and Presses and of Small Press Distribution, suggests that the difference between Amazon and brick-and-mortar bookstores is most evident in how they market books: “I think even people at Amazon would say that it’s essentially a widget seller that happens to have begun by focusing on books. Many people, like me, will say you can’t sell a book the same way you sell a can of soup.”
At the heart of the soup-can analogy are the algorithms that Amazon uses to “recommend” books to customers. Most customers aren’t aware that the personalized book recommendations they receive are a result of paid promotions, not just purchase-derived data. This is frustrating for publishers who want their books to be judged on their merits. “I think their twisted algorithms that point you toward bestsellers instead of books that you might actually like [are] a shame,” Gavin Grant, cofounder of Small Beer Press, laments.
As a reader, reviews don’t do shit for helping me find a book to read. Personally, I tend to ignore consumer book reviews (and some professional reviews as well) when looking for something to read. I have a few trusted sources, but I also rely heavily on scanning the first few pages—or sometimes a random spot in the middle—to get a sense of whether the book is right for me at the time.
Even that takes a lot of time, and I sometimes make bad picks. What I want is a more reliable way of knowing a book’s quality relative to other books I’ve read and enjoyed.
If only there were a way to measure a text’s relevance and quality. Some sort of classifier, let’s say, that weeds out undesirable text and narrows my choices down for me. After all, making a decision in the face of too many choices is overwhelming. And in today’s book market we have a lot of choices.
We’re at a unique point in history. More text today is being published in electronic format than ever before. We shouldn’t have to rely solely on other people’s opinions (although we shouldn’t ignore them wholesale either) when it comes to finding a good book.
Right now, there’s a little magic at work for you helping you decide which emails are worth reading and which are probably junk. It’s a spam filter, and nearly every hosted email service and every email client has on.
With so many books in electronic format, why on earth aren’t we using a similar approach for classifying? I won’t get into technical details (cause this ain’t a technical post), but trust me when I say this is not only possible, but is also now practical. With so many modern works available as e-books, implementing a preference filter is well within reach. In fact, I wouldn’t be surprised if Google Books was headed this way. After all, what’s Google doing with all those scanned books? Textual analysis.
There’s a post up at the Wall Street Journal that speaks to the use of algorithms to sort and classify creative works. It briefly touches on applying algorithms to text but focuses more on using those algorithms to generate writing and only mentions using algorithms to grade text. I think it misses a huge potential for Natural Language Processing (NLP).
Using algorithms such as Naive Bayes (typical SPAM filter), we can let a consumer categorize books they’ve read and use NLP to classify unread books based on those categories. For instance, say I enjoy reading Science Fiction, Horror, and Romance. I could create a list for each genre, and my bookseller (say, Amazon) could then let me search their collection for books that fit into my personalized categories.
The WSJ article also mentions the use of NLP for grading papers, and claims the current systems can grade as well as any human. So we also have tools at hand to help us find not only books that fit into our categories, but books that are written well. This ranges from simple grammar and spell checking (which self-publishers may either fail to do or do poorly) to Readability formulas.
Here’s where you say, “What? Books are an art form! There’s no way a computer could understand art well enough to distinguish the good from the bad.”
Yeah, okay. True, a computer may never fully understand the nuances of an art form, but all art is based on fundamental rules. Music, movies, books, painting—all have basic elements the artist uses in composition, and those basic elements can be quantified.
And you know what? We already use algorithms to help pick music (Pandora, Ping) and movies (Netflix, Clicker). And as mentioned above, Amazon uses algorithms (although allegedly inappropriately) to recommend books.
I could drag on for hours about this, in part because it blends two things I’m passionate about (books and technology).
I’m thinking about using Kickstarter to fund a book recommendation service based on a combination of NLP algorithms, with initial focus on genre fiction. The service wouldn’t take the place of human reviews, but could provide a sort of litmus test by which reviews could be tempered. Setting this up would take a lot of work, and there’s no way I could do this on my own. So before I go any farther, I’d like to ask a few questions to help me gauge if this is worth my time:
As a reader, would you use a service for book recommendations (likely for free) knowing the service used only algorithms?
As a gatekeeper (agent / editor / publisher), would you be interested in a service to classify your slush pile, comparing submissions to works you’ve previously published (or works you select) and scoring them for grammar, spelling, and readability? Would you be willing to pay a small fee?
As a writer, would you be interested in an automated service to help you locate potential markets for your work? Would you be willing to pay a small fee?
Have at it. Don’t feel the need to answer the questions. General comments are more than welcome, too.