4 ms·
This is fascinating. For those that want to save time reading, the author used a text classifier and had a 70% accuracy rate predicting if a book was published
by jedberg 7y ago
This is fascinating. For those that want to save time reading, the author used a text classifier and had a 70% accuracy rate predicting if a book was published by a conglomerate or an independent publisher.
Which means there is a distinct difference between the type of content that indies will publish vs conglomerates.
Whether this has an effect on literature is what the rest of the article is about.
- vages 7y agoWas also surprised by the machine learning twist. A good read, with a promising ending for new research.
- Animats 7y agoAll his "conglomerate" texts are from Random House. He may have simply measured Random House's editorial style. Other publishers may differ. Random House has a house style, and guidelines. Here are the guidelines for their Penguin division.[1] Random House accepts submissions only through established literary agents.[2] So they have an organized filtering process. [1] http://authornews.penguinrandomhouse.com/what-our-editors-look-for-on-an-opening-page/ http://authornews.penguinrandomhouse.com/what-our-editors-lo... [2] http://www.penguinrandomhouse.biz/manuscripts/ http://www.penguinrandomhouse.biz/manuscripts/
- saghm 7y ago> the author used a text classifier and had a 70% accuracy rate predicting if a book was published by a conglomerate or an independent publisher Genuine question: is 70% actually a very successful rate? It's hard for me to tell without more information; if 70% (or 30%) of books are published by independent publishers, then 70% would be the same rate as just guessing the more common one every time.
- c3534l 7y ago70% feels pretty awful to me when you don't even know what it is you're measuring.
- sterkekoffie 7y agoIt depends on your definition of success but 70% accuracy for classification problems with 50/50 odds is at least useful. In this case, "guessing the more common one" would not be a problem as the data were evenly split between indie and RH.
- cma 7y agoCould it be something like a classifier having success at telling if someone is a professional NBA player vs a club basketball player, or is there something more to it?