Recommendation Exploration, Pt. 2
Posted by: Neil Hollands
A week ago I explored the best techniques for recommending books to others in a book group setting. Today I’d like to look at how well various institutions do in making suggestions, particularly recommendations based on a reader’s expressed preferences.
At Williamsburg Regional Library, we’ve made our “Looking for a Good Book” service a point of pride. Our patrons can complete a reading profile and within a week get back a personalized list of reading suggestions. It was innovative when we started it 9 years ago and has since been copied by dozens of libraries across the country. It remains a staple of the services that we offer to readers.
Of course your library may not offer this service or you may be the kind of reader, like me, who is always looking for new perspectives on which book to try next. Online booksellers, social media sites, and other bookish places on the internet offer book recommendations, but are they any good?
Whichbook is one of the most venerable online recommendation services. At this site, users mark preferences that indicate the tone of their preferred reading on sliding scales, or select character, plot, or setting options. Whichbook returns a short annotated list of titles. It’s a cool concept, similar to what librarians do with reading profiles at WRL, but Whichbook’s form offers less detail and results come from a finite set of books, many of which are by English authors. The results are fun but limited and reflect typical problems with taking humans out of the reading suggestion equation: an automated service is only as good as the detail of data being input behind the scenes and the frequency with which that data is updated. Automated services also tend to overemphasize new titles at the expense of all the lovely backlog that a good library contains.
For years, Amazon offered recommendations based on the books that one had rated and purchased on the site, but the results were never good, and a few years ago, just before their focus turned to Kindle-mongering, the service was quietly dropped (as was the ability to rate items without posting a review.) Now Amazon relies on other means of suggestion, mostly tied to which products have most often been purchased by people who bought a particular item. The results can be interesting, but usually begin with the terribly obvious: almost every other title by the same author and big, recent hits from the same genre. It’s a good indication that relying mostly on sales data to provide reading suggestions is of limited value.
GoodReads, the reading social media site, recently enhanced its recommendation options. There were always the reviews of others, lists, tags, and polls to provide suggestions (and such options are available on competitors like Shelfari and LibraryThing as well.) GoodReads now has an additional service, providing recommendations based on books added to one’s list and the ratings given to those books. These are compared to the input of other readers of the same titles and the rating scores of other books with similar genres or subject matter. GoodReads then provides fifty suggestions at a time in each genre or subject that you frequently read. The results are both more apt and more diverse than those on Amazon, with many older, even out-of-print titles coming back into the mix. You can indicate interest or disinterest in these results to further fine tune suggestions. It’s the best automated suggestion service that I’ve encountered, although still not quite as strong as recommendations from a well-trained readers’ advisor. Still, it’s fast, free, infinitely repeatable, and tied to all those great reviews written by other readers.
In the end, any kind of artificial intelligence is only as good as the logarithms running in the background and the detail of the metadata from which the data is drawn. As reader data continues to aggregate and intelligent programmers turn more attention to the problem, the results are bound to keep getting better, but for now, you’ll still benefit most from the artful recommendations of voracious readers who know how to elicit your reading preferences skillfully and communicate suggestions gracefully.