LibraryThing Coverage
Microsoft Searches for Relevance in a Google World

BookHints on

AbebookslogoThe folks at recently introduced a nifty recommendation service and they're calling it BookHints.  The underlying technology for this feature is provided by...drum roll please...LibraryThing, the subject of a post yesterday.

The AbeBooks announcement notes that only about 10% of their list currently offers BookHints, so I got curious to see what they recommend.  It took me a few tries, but I found a book from my group that is in the program: Lifehacker, by Gina Trapani.  If you click on this link you'll see a BookHints link just below the cover image.  Click on that link and you'll see the 6 titles BookHint recommends.

You might be wondering why this feature is any better than Amazon's "Customers who bought this item also bought" list.  To me, it's the difference between customer purchases and customer recommendations.  The former is interesting but doesn't have a direct correlation to customer satisfaction; I might buy two books at the same time, but I might only wind up liking one of them.

If AbeBooks and LibraryThing truly want to offer a compelling feature with BookHints, they need to limit their title displays to only those books with the highest reader ratings.  I'm not sure this is happening today with BookHints, mostly because the implementation specifics I've seen aren't that detailed.  I'll see if I can get someone from either/both organizations to comment on this.

My only other beef with this service has to do with the user interface.  Rather than forcing me to click through to a new page to see the BookHint recommendations, why not just do a pop-up window which appears on top of my main window till I move my cursor away?  I'm seeing more and more pop-ups like this and they're not the annoying ones of the 1990's -- this truly is a useful interface option and lends itself to this sort of web page feature.


Richard Davies

Dear Joe, members rarely catalog books that they dislike. The average rating for a book on LibraryThing is almost four out of five stars. Therefore we are confident that the recommendations are books that were liked.

The fundamental point behind BookHints is that the recommendations are based on books that are owned, read and liked by like-minded people. They are not simply based on purchasing habits. For instance, I bought Charlotte's Web for my niece's birthday at the weekend but I'm not going to read it.

Richard Davies

Joe Wikert

Hi Richard. Thanks so much for providing more information on BookHints. You make a great point about how there's often no correlation between book A and book B in the same transaction. It's also interesting to hear that the average book on LibraryThing is rated so high. I'm finding that if I lose interest in a book I'm reading I not only don't finish it, but I also don't review it on my blog and I don't add it to my LibraryThing list. I guess others tend to do the same.

Tim Spalding

Hey Joe. I hope you won't mind if I jump in. Recommender systems have become something of a passion--that's why LibraryThing has *five* visible systems now, not counting our "unsuggester." All these together are turned into the "combined" recommendations on the work pages, which turn into Bookhints.

I've played with ratings algorithms a lot. They're strangely useless. On their own, they make very erratic recommendations. As a contributing factor, they don't add much.

I think the reasons are:

1. A minority of books have been rated.
2. The average is about 3.8 stars. In itself that's not bad, but it makes the signal weaker, like trying to figure out who are the smart kids at Harvard, where everyone gets As.
3. I don't know where I read it, but someone looked at star ratings and sales on Amazon. The strongest correlation between high sales and stars was the number of *one-star ratings*. I think that's because a really popular book spreads beyond its natural readership. People buy it to be part of the conversation. And it becomes a sort of target. That messes with things.
4. Even if you don't like a book in your LibraryThing library, you still bought it (or, occasionally, read it from a library, a friend, or etc.) LibraryThing is not going to recommend THAT book again to you, so it's about whether the book has some affiliation with books you might like. People buy books they don't like certainly, but I'm not sure they often buy books of a *sort* they don't like. In a way, the books in your library represent what you want to like as much as what you do like. It's possible he books you had high expectations for, but didn't end up liking, are just as valuable for recommender systems as anything else.

Those are my speculations anyway. They data isn't very useful, but my guesses as to why may well be false.

All this said, I may revisit the problem. It's possible the very lowest end could be useful. Ditto real favorites. And even if they don't make the best recommendation system, ratings have their uses. The Zeitgeist Books page has an interesting list—books people disagree on. This is calculated from the standard deviation of ratings. The winner—the Book of Mormom—is a perfect example: Nobody is in the middle there. You either love it or hate it.


Joe Wikert

Hi Tim. Thanks for laying all this out for us. And congrats to you for all the interest currently swirling around LibraryThing -- as you can see, I proudly feature your widget on my blog! I used to have the Shelfari one as well, but dropped it, mostly for one simple reason: It didn't offer the randomize feature that LibraryThing does. I love it that different books from my library pop up each time someone else visits my blog. Great job, and thanks again for making it free...although as I noted in my earlier post, you'll be getting money from me soon enough!

Verify your Comment

Previewing your Comment

This is only a preview. Your comment has not yet been posted.

Your comment could not be posted. Error type:
Your comment has been saved. Comments are moderated and will not appear until approved by the author. Post another comment

The letters and numbers you entered did not match the image. Please try again.

As a final step before posting your comment, enter the letters and numbers you see in the image below. This prevents automated programs from posting comments.

Having trouble reading this image? View an alternate.


Post a comment

Comments are moderated, and will not appear until the author has approved them.

Your Information

(Name and email address are required. Email address will not be displayed with the comment.)