Fair sequential group recommendations

Recommender systems have been incorporated in our everyday life; from music to health recommendations, recommender systems have enhanced the users' experience. At the same time, with the expansion of social media, it is now easier than ever to form groups of people. As such, group recommenders have become more popular. Often, we consider the interaction between a group and the recommender system as a stand-alone process; the group requests some suggestions from the system and the system answers without any considerations to past interactions. A more realistic scenario is for a system to require access to history logs, and take them into account when recommending items for a group of users. Not only what items the system previously had recommended, but also, how well were these items received by the members of the group. In this work, we propose a sequential group recommender, which is aware of the past interactions of the group with the system. We introduce the notion of satisfaction that describes how relevant are the recommended items to each member of the group. We utilize satisfaction in a novel aggregation method that achieves to make our model fair for all members of the group. We show with experimental results that the typical group recommendation approaches are substandard to our proposed method.