Collaborative Research: CCRI: Medium: A Research News Recommender Infrastructure with Live Users for Algorithm and Interface Experimentation

The goal of this project is to develop a testing infrastructure for news recommender systems. One important task is to develop text mining algorithms to automatically classify news stories into types that can be used to establish a profile of what content interests each reader has and then identify stories in the future that cover topics of interest. We will also be working on improving recommender system algorithms. Some of the specific improvements include creating “bundles” of story recommendations in the form of a newsletter or landing page. While the stories should be of interest to a particular user, there are many other considerations. The bundles should have a diversity of topics in them and a mixture of topics that the user follows as well as surprise articles that the user was not expecting. They must also consider partisan leanings of news stories so as not to create filter bubbles. Another important task will be to use reinforcement learning to learn the preferences of new users and help them explore the breadth of content available.