This is interesting:

BookLamp.org is a system for matching readers to books through an analysis of writing styles, similar to the way that Pandora.com matches music lovers to new music. Do you like Stephen King's It, but thought it was too long? The technology behind BookLamp allows you to find books that are written with a similar tone, tense, perspective, action level, description level, and dialog level, while at the same time allowing you to specify details like... half the length. It's impervious to outside influences - like advertising - that impact socially driven recommendation systems, and isn't reliant on a large user base to work.

The website has a video that explains the ideas behind it...

I've talked before about book recommendations, but this is the first one I've heard of that analyzes writing style and uses it as the basis for the recommendation. So I signed up and took it for a test drive...

There are only 179 books in the database to date - not a huge number but something to get a flavor for how it works. But here's the cool news: They are all science fiction titles. Sweet!

I stared with a book I've read: The Naked Sun by Isaac Asimov. The resulting recommendations are:

• The Beyond by Jean and Jeff Sutton (56% match)
• Lord of the Stars by Jean and Jeff Sutton (55% match)
• The Rock Rats by Ben Bova (55% match)
• A Talent for War by Jack McDevitt (52% match)

I never heard of the first two and haven't read the other two, so I'm not sure of how well it's done, no how good a match it would be at 55% anyway.

I'm going to try again, but first I note that the results also include some books statistics, including number of words, a piece of information that is otherwise unknown to me. It's nice to get a feel for word count, I guess, especially when some author blogs talk about story sizes. (FYI: The Naked Sun clocks in at 67,239 words.) There is also a button enticingly labeled "Show Me the Graphs from this book". I press it. Up pops a semi-transparent window with more stats...including: scene count (71), pacing (7), density (4), action (9), description (5), and dialog (9). These are the test points used to compare other texts, though for me they currently lack any frame of reference. There is also a graph where these 5 characteristics are plotted (you are able to enable each plot individually on the same graph. Also helpful: you can click on the characteristic labels to see what Book Lamp means by these terms; for example, what a high pacing rating means.

My next book was Rogue Moon, by Algis Budrys. Recommendations here included The Web Between the Worlds by Charles Sheffield (77%), Manhattan Transfer by John E. Stith (76%), and Bright Messengers by Gentry Lee (78%). I'm not sure why results are not displayed in high-to-low percentage order, seems like they should be. Again these are books I haven't read, so I cannot speak about their success.

I tried some other titles, but I'm not going to list the details -- being a flash application makes it too time consuming. (No cut and paste for you!) But I will note some data:

• The best match for David Brin's The Postman (out of 8 recommendations) is Dinosaur Planet by Anne McCaffrey (58% match).
• The best match for David Drake's The Far Side of the Stars (out of 5 recommendations) is The Burning World by J.G. Ballard (58% match).
• The best match for David Drake's With the Lightnings (out of 2 recommendations) is The Rising by James Doohan (77% match). Odd, since The Far Side of the Stars is part of the same series and both are in their database. Or, are they drastically different writing styles? I would think books by the same author in the same series would match better. (The other match was Savages of Gor by John Norman.)
• The best match for David Gerrold's Jumping Off the Planet (out of 3 recommendations) is The Stainless Steel Rat Joins the Circus by Harry Harrison (58% match).
• The best match for Robert J. Sawyer's Calculating God (out of 8 recommendations) is a tie! The Mindwarpers by Eric Frank Russell and Forward the Foundation by Isaac Asimov both get an 80% match, the highest number I've seen so far.

It's hard to tell how well this site is working since (1) Their database is relatively small, and (2) I have not read many of the titles that are in their database. (Maybe you folks out there can attest to their success?) I still like the idea, though, and am interested in seeing where this leads.

Comments (1) | PermaLink | Category: Web Sites
Posted by John DeNardo at Saturday March 22, 2008 at 12:33 AM
© Saturday March 22, 2008 at 12:33 AM SF Signal

Post a Comment

Name:
Email Address:	(Will not be displayed)
URL:
Remember me?	Yes No
Comments: