Multi-Agent Meeting Scheduling with Time Preferences
Khaled Al Wahedi, Osama Mussawar
جامعة خليفة · الامارات
The problem of meeting scheduling, in its general outlines, is a typical problem of group decision. The main question, addressed in group decision is, given the individual (private) preferences over alternative mutual decisions, what should the mutual decision be. Meeting scheduling is an inherently difficult and time consuming task that requires multiple parties to interact and consult their various calendars in order to agree upon a common time. In real world applications, users usually have conflicting preferences which makes scheduling a meeting generally more complex in that it attempts to satisfy the preferences of all participants. All participants can agree to a schedule in which a portion of their preferences are not satisfied, since preferences are regarded in terms of the relative importance of daily activities. Usually, there is no a-priori knowledge of all the possible constraints, local or global, that may apply when scheduling a meeting. In addition to this there is an overhead in terms of time and effort involved in scheduling a meeting that can sometime make it impossible to schedule an event. This necessitates the automation of this process, which can also lead to more efficient and satisfying schedules. We formulate and solve the scheduling problem from the perspective of making a group decision, optimal in some sense, about the time of meeting. Due to its distributed and dynamic nature, meeting scheduling is modeled as Multiagent System. Three different approaches are developed to schedule a meeting, varying in terms of information structure, the decision mechanism, size of the set of agents and that of outcomes. The first approach is based on central agent responsible for decision making using optimization. The second approach requires the agent to interact and negotiate with each other under the specified protocol. In third approach, agents follow a social decision mechanism under the influence of their preferences. The decision maker agents are also modeled, which specify how they interact and behave in multiagent decision mechanisms. Certain human behaviors are embedded in the design of agents, which also enables them to converge. Development of effective algorithms to automate the meeting scheduling process is also part of this work. We develop a simulation framework in which agents are created and they interact with each other to schedule. The proposed mechanisms are analyzed and simulation results are presented.