Project Alexandria – A graph based book exploration and recommendation engine

by ankit_tandon • 11 years ago

29 9 11 years ago

oinksoft• 11 years ago

It took a few tries to find a non-classic book I've read and own. I was surprised by the neighbors/recommendations for Henri Troyat's Tolstoy:

- George Orwell's 1984, a "Bloom's Guides" edition (Cliff's Notes-looking thing)

- Benjamin Franklin, a graphic-novel format biography of the statesman and scientist

- Gran Pionero de Beísbol, a graphic-novel format biography of Jackie Robison

- Wilma Rudolph, a graphic-novel format biography of the Olympian runner

- Martin and John, "with a frankness about sexuality rarely seen in contemporary fiction ... a haunting, daring, and heartrending debut novel that speaks of gay life"

Those results don't seem to reflect the stated methodology in the "About" section... the only commonality is that three of the five are biographies. I expected more things to do with Russian literature, or literary biographies contemporary with Troyat's.

I'm interested in the project's implementation, it's too bad it isn't open source.

danpalmer• 11 years ago

Took me a few popular authors to find one that was in the database, and when I finally tried the book I'm reading, the only recommendation it said it was like was way off - The Devil in White City has no relation that I can see to The Lost Symbol.

ankit_tandon• 11 years ago

Add a ":" before your search for author search. Sorry should have made that clearer

stared• 11 years ago

It would be great if the graph was larger. Or rather - any non-trivial graph, with diameter > 2 and interconnections (rather just star of 6 nodes).

gilleain• 11 years ago

If you click on a leaf node, it walks the graph, expanding as it goes. Managed to find a triangle at one point. I wonder what the edges really are - some similarity measure, but what exactly?

odabaxok• 11 years ago

If you hover over an edge you will see what similarity measure it uses.

sundarurfriend2• 11 years ago

Interesting project, and the visual display is appealing, but the results don't seem to be any better than from the recommendation engines of, for eg., Goodreads. In many cases, the other nodes end up being mostly other books by the same author, with one or two books by different authors in the same genre.

One thing that could improve this as an exploration and discovery engine is an option to disallow books by the same author as the original we input. That would allow me to, for eg., check out what books the engine thinks are similar to 'Thief of Time', without the result being just a bunch of other Discworld books (which is not really useful or new information).

Also, why so few related nodes (five) in the result? Is this a constraint placed explicitly or just a feature of the algorithm? Having more books returned would also help increase the probability that the user finds something actually useful and previously unknown to them.

sundarurfriend2• 11 years ago

Another interesting possibility would be allowing the user to enter two books, and then showing what the engine thinks the routes from one to the other are.

smartial_arts• 11 years ago

Not sure how the search suggestions are meant to work, but when typing "Drunkard's Walk" I can see the book at the top of the result's list, although when typing the author's name, "Leonard Mlodinow" I get no matches for his books at all.

Project Alexandria – A graph based book exploration and recommendation engine

Related Stories