Towards benevolent sales assistants in retailing scenarios

Abstract

Non-collaborative dialogues like sales dialogues are characterized by congruent intentions, i.e. intentions that are agreed by dialogue partners, and conflicting intentions. We will refer to these intentional structures as mixed intention sets. In this paper, we will investigate dialogue systems that are benevolent towards a dialogue partner, i.e. benevolent agents try to find a fair balance between partner intentions and agent intentions in particular with respect to conflicting intentions. For the class of question-answering dialogues, we propose a model for the intelligent generation of answers considering mixed intention sets and demonstrate its application in the retailing domain in form of a benevolent sales assistant (BSA). BSA processes mixed intention sets in a strategic way by means of a game-theoretical equilibrium approach to find a fair balance between intentions of dialogue partners. We evaluated the BSA by a run-time analysis of 500 simulated sales dialogues between customers and retailers and show how the sales assistant strategically generates answers considering mixed intention sets in retailing scenarios.