Toward an Optimization of Elicitation Timings Considering Elicitation Costs and User Privacy for Personal Assistant Agents

In this paper, we present an overview of an optimization that a personal assistant agent learns timings to elicit preferences from their stakeholders while negotiating with other agents to execute each task cooperatively. We consider a negotiating agent that represents an associated stakeholder with only limited information about user preferences. To avoid bothering users and exposing their privacy to ask about their preferences, we present a mechanism that allow personal assistant agents to learn the policy to elicit preferences from their stakeholders using a Q-Learning based approach.