Visualizations and Explanations for Sequential Group Recommendations

Abstract

The popularity of sequential recommendations is on the rise these days. It is important for the system not to treat each round of recommendations as an independent activity; rather, it should store information about previous encounters. More and more people are creating groups for activities, which makes group recommendation systems more popular. It frequently happens, however, that recommenders are unable to find the most useful data pieces. This flaw is addressed by explaining why specific suggestions are given. This work proposes visualizations for recommendations generated by SQUIRREL, A Framework for Sequential Group Recommendations through Reinforcement Learning. We explored three why questions using the 20M MovieLens dataset. Explanations rely on the aggregation method used for the last iteration for a particular group, combined with single-user and group recommendations. The Graphical User Interface framework incorporates visualizations and explanations. We have used three test cases and are able to provide explanations personalized for each group.