In both special-purpose applications such as an on-line course text, a travel recommender, TV-guide or an art recommender, and general-purpose websites such as an on-line encyclopedia or even Web search, end-users nowadays expect a personalized approach. The adaptive technology that generates this personalization has traditionally made use of the brains of domain experts to make decisions what to show or recommend, which hyperlinks to offer and how to adapt the presentation to the user’s device, location and other context information that is available.

The expertise can be can be completely hand-crafted by the expert, for instance to define a structure of prerequisites within an on-line course, or can be made more generally usable, like a semantic database about art styles, techniques and artists, about geographic locations, etc. so that the adaptation can reason over these data and the user’s actions to decide what to recommend. This leads to what is called content-based recommendations or adaptation. At the TU/e we have in the past realized several examples of expert-driven adaptive applications, culminating in the world’s first adaptive PhD thesis in 2012.

But as the information space and the diversity of its users grow the expert-driven adaptation approach becomes prohibitively expensive and time consuming. Hence the new challenge of data-driven adaptation. The first recommender systems using collaborative filtering are already quite old and so is the knowledge of their limitations and the (stupid) mistakes they can make. The new research challenge is to use data-driven adaptation in combination with expert knowledge to make adaptive technology that continuously improves its adaptation based on data gathered about its usage. In science-fiction this inevitably leads to the decision that the optimum is to eliminate humans from the world. At the TU/e we will research possibilities to reach a different optimum. 

Public visé :

Séminaire interne interne de l'équipe SILEX mais ouvert à tous.