Programmers across the world are rushing to apply machine-learning to pretty much everything, from bicycle helmets to refrigerators. Most recently, a team of researchers in Japan has semi-successfully used it to confirm that the homogenous designs of books within a genre is a pretty effective marketing technique.
Brian Kenji Iwana and Seiichi Uchida at Kyushu University utilized a deep neural network to gauge whether a computer could correctly guess the genre of a book based on its cover. They fed the covers and genres of around 100,000 books found on Amazon to the network, training it to recognize patterns in the images associated with each of the twenty genres used in the dataset. According to MIT Technology Review, “the algorithm listed the correct genre in its top three choices over 40 percent of the time and found the exact genre more than 20 percent of the time.”
Until a similar study is done to gauge human beings’ ability to correctly ascribe genres to books using only their cover images, it will remain unclear whether the computer’s occasional success is better or worse than the strategies our brains employ.
The similarities between book covers within genres has, however, been well-documented–particularly within romance–indicating that humans are pretty good at the task, and publishers are good at taking advantage of it. It makes sense: repetition is a common facet of any marketing campaign or branding initiative. Publishers try to capitalize on our responsiveness to it by reinforcing the association we have with, say, cursive text and romance, or an image of rolling hills and travel.
Read more here: www.forbes.com/sites/ellenduffer/2016/11/19/computer-trained-to-guess-book-genre-using-only-cover-image/