On Radio-Canada’s Tou.tv web platform, users can view content on demand, at their own pace and according to their tastes, provided they take the time to explore the Tou.tv website and find a show they like. This task can be tedious, however, and can cause users to miss potentially interesting content. Recommendations would most certainly be welcome!
This is the subject of a research partnership between Radio-Canada and Laurent Charlin, assistant professor in the Department of Decision Sciences at HEC Montréal. Radio-Canada provided the researcher with anonymous data from Tou.tv users’ viewing histories. It is a rich source of information for finding out not only what content users viewed, but also whether or not they liked it based on how long they watched. Using this data, the recommendation algorithm works by matching viewing histories, in other words by associating users who share similar tastes. To make recommendations to a user, it analyzes Tou.tv contents by filtering them through the histories of similar users. Recommendations will consist of the content enjoyed by similar users that the targeted user has not yet watched.
We’re interested in content-recommendation algorithms because the Tou.tv catalogue is large enough to personalize the user experience based on what they’ve watched.
Senior Director, Business Intelligence, Radio-Canada
This strategy could, however, lock the user into a bubble and go against Radio-Canada’s mission. One possible solution is to add some random content not identified by the algorithm to the recommendations to allow the user to discover new programs. If the user chooses to watch them, the system will then incorporate their content history.
Obviously, in addition to improving users’ experience, the recommendation system aims to extend their time on Tou.tv.
We find Radio-Canada’s data interesting because it has features that other data sets don’t.
Assistant Professor, HEC Montréal