Kibitz : a framework for creating recommender systems

Abstract

Recommender systems are one of the most vital and ubiquitous parts of the modern web. They are used by many major internet services such as Facebook, Google, and Amazon. However, there is a wealth of content and data that remains untapped by mainstream commercial recommender systems. We have designed and implemented Kibitz, a framework that allows anyone to create a recommender system on top of an arbitrary collection of items. We have developed a web application that facilitates the creation, customization and deployment of standalone websites for browsing and rating items as well as receiving item recommendations. We have also created a set of libraries for embedding rating and recommendation functionality into other websites. Partnering with local bookstores, we evaluated the process of using Kibitz to build recommender systems for their communities.