Sandip Sen

Department of Mathematical & Computer Sciences
University of Tulsa
600 South College Avenue
Tulsa, OK 74104-3189
Phone: (918) 631-2985
Fax: (918) 631-3077
E-mail: sandip@kolkata.mcs.utulsa.edu
URL: http://euler.mcs.utulsa.edu/~sandip/



"RIA: Designing and Implementing a Distributed Meeting Scheduler"


The PI received an NSF Research Initiation Award in 1994 from the ITO subdivision of the IRIS program of the CISE Directorate. The PI believes that routine, time-consuming, and repetitive information processing tasks performed by employees in organizations can be potentially automated. Possible benefits from automation include relieving employees to concentrate on relatively fruitful and productive jobs, and in improving the quality of work by eliminating human errors that are inadvertently introduced due to the tedious nature of work. In this proposal, the PI proposed to investigate such a candidate task for automation: that of scheduling meetings between individuals. The goal of the proposal was to develop the mechanisms required to effectively schedules meetings without compromising user preferences and constraints. The research priority of the PI was to develop general distributed negotiation mechanisms between automated agents that can trade-off conflicting interests.

Results From Research

Using User Preferences:
One of the major scientific components of this research has involved the development of a structure, formal mechanism for representing and using user preferences for automating the negotiation for scheduling meetings. This aspect of our scheduler clearly differentiates our proposed system from other commercially available products. User preferences for meeting topics, proposed meeting length, host of the meeting, invitees to the meeting, etc. are used to assign a priority for an alternative for a meeting. For example, user A may be obligated to attend meetings hosted by user B, but not if B was only an invitee. Each user assigns a value between 0 and 1 for each option of each dimension. The user also specifies a minimum threshold for that dimension. If the value of the option falls below the minimum criteria, then for that dimension the user would prefer not meeting over meeting. The user is also responsible for rating each dimension against each other, e.g., who is hosting the meeting may be more important than whether the meeting is being held in the morning or in the afternoon.

We assign votes to each preference dimension in proportion to their weight against each other. Each meeting proposal can then be voted on by the preferences as to whether the user wishes to meet given the meeting criteria. For each option above the dimension's threshold, yes votes in proportion to the weight of that dimension are recorded, and for each option below the dimension's threshold, no votes in proportion to the weight of that dimension are recorded. We adapt the Black's voting rule which guarantees both Pareto and Condorcet winner criterion. Our adaptation is unique because it allows easy use and comprehensibility on the part of the user, without sacrificing the formal properties guaranteed by the underlying bidding mechanisms. Our approach is a comprehensive one that can be used to develop other personal information processing agents that can share some of the routine information processing tasks performed by users in organizations. The system will provide some default preferences which each user can modify according to his/her liking. We also anticipate a learning mechanism to modify these preferences, which will be particularly useful for naive users who do not modify the preferences from their default values.

System Implementation:
In designing the architecture of the implemented distributed meeting scheduler, we had to concentrate on designing a system that would be readily extensible to accommodate functionalities that may be added to the system at a later date. The single major design decision we adopted was to use the e-mail system as the communicating paradigm between different meeting scheduling agents. Following the same design philosophy, we chose to use the calendar management system supplied by most operating systems to store and retrieve the calendar information of the associated user. Each of the modules in the architecture can be briefly described as follows:



Publications Resulting Directly From This NSF Award:
  1. T. Haynes, S. Sen, N. Arora, \& R. Nadella, ``An automated meeting scheduling system that utilizes user preferences,'' in Proc. of the First International Conference on AUTONOMOUS AGENTS (Agents '97) held in Marina del Rey, CA between February 5-8, 1997.

  2. S. Sen, ``An automated distributed meeting scheduler,'' accepted for publication in IEEE Expert.

  3. S. Sen and E.H. Durfee, ``A Formal Study of Distributed Meeting Scheduling,'' accepted for publication in Group Decision and Negotiation Support Systems.

  4. S. Sen and E.H. Durfee, ``A Contracting Model for Flexible Distributed Scheduling,'' Annals of Operations Research, volume 65, pages 195-222, 1996.

  5. S. Sen and E.H. Durfee, ``Unsupervised Surrogate Agents and Search Bias Change in Flexible Distributed Scheduling,'' in Proc. of the First International Conference on Multiagent Systems (pages 336-343), San Francisco, California, June, 1995.






Return to ITO Workshop Abstracts

Return to ITO Workshop Home Page