Patrick Brézillon, University Paris VI:
"Explaining for Contextualizing and Contextualizing for Explaining"
"The presentation aims to show that the time to renew with explanation generation is now coming. The first part will be back in the past in order to understand goals and objectives in the eighties and the severe limits due to a lack of operational concepts, tools and technology (it was the time of the early expert systems and the so-called knowledge-based systems). The lessons learned from this time point out the need to introduce the user (i.e. the explainee) in the loop of the explanation generation as a partner. Another important lesson was that if additional knowledge was required, it was not domain knowledge only, but a real context management. In the second part of the presentation, we discuss a set of new types of explanation in a context-based formalism called contextual graphs, and the need to consider explanation generation with knowledge acquisition and learning as intrinsic parts of any task at hand. We begin by presenting the context-based formalism of representation and after an explanation typology that can be established, thanks to Contextual Graphs. However, other context-based approaches can be consider too." Thomas Roth-Berghofer, DFKI GmbH / TU Kaiserslautern:
"Revealing the Magic of Product Recommendation"
(Joined session with ECAI-Workshop on Recommender Systems W19)
"Recommender systems have a certain magical quality. Users look for interesting products or useful information and, often miraculously, get some suggestions alongside their browsing results. In this scenario, recommender systems take on a role we usually ascribe to colleagues and friends who help us choose one product over an other. Their advice may be based on content---the features of this product seem to have advantages over the features of that product—or the recommendations are community-based—others bought this product, too. But where we can ask a human about their advice, on what they based their recommendation, recommendation systems not necessarily are able to give us this information. They do not explain their reasoning and how they came up with the suggested solution, although recently considerable research has been undertaken to remedy this situation."
|
