Kobo is the second largest digital bookstore in the world and they have over five million titles in their catalog. One of the big challenges is being able to accurately recommend titles to individual users based on their reading habits. Kobo has addressed this problem by leveraging big data and machine learning.
“Modern recommendation systems, no matter how technically sophisticated, are based on a very simple foundational idea: similar people like similar things,” explains Kobo’s Director of Big Data, Darius Braziunas. “If you see a bookshelf and recognize some of the authors and books on it, it is likely you will enjoy reading the other books as well.”
Kobo analyzes the titles in a user’s library, and automatically extracts the main topics of interest — using unsupervised machine learning techniques — and creates a weighting of these interests based on their relative importance to the user. A user might be interested in both science fiction and biographies, but if the last few books she has read are science fiction, the recommendation system will assign a higher value to that topic, leading the user to receive more sci-fi recommendations than biography.
“Our goal is to make the entire website personalized, to customize to individual interests, whatever those may be.… To do this we take into account a variety of data points — offering what we hope is a richer, more relevant list of suggested reads,” said Braziunas.
The recommendations that Kobo provides are found on personalized lists that live on the homepage, “Next Read” emails sent when a user is close to finishing their current book, related books (“people who read this also enjoyed…”) on each product page, on the company’s line of e-readers and apps for Android/IOS.