Utility Elicitation During Negotiation with Practical Elicitation Strategies

Automatic negotiation is gaining more interest recently thanks to the wider deployment of intelligent systems and the need for them to cooperate/compete on behalf of their users. A central assumption of most autonomous negotiation agents is that the utility function of the user is perfectly known to the agent. That is an often unmet assumption in real situations. Utility elicitation is the process of learning about the utility function of the user incrementally and has a long history in decision support research. Recently, some utility elicitation systems capable of incrementally eliciting the utility function of the user during the negotiation were presented. This work expands this body of research by optimizing the elicitation algorithm to realistic elicitation strategies. The proposed method extends the optimal elicitation algorithm to the - practical - case where queries to the user only reduce the uncertainty in the utility function without removing it completely. Extensive evaluation shows that the proposed extension outperforms two state-of-the-art elicitation algorithms and several baseline alternatives.