UPM Institutional Repository

Neural Network Preference Learning Approaches For Improving Agent-Based Meeting Scheduling Problems


Citation

Tang, En Lai (2007) Neural Network Preference Learning Approaches For Improving Agent-Based Meeting Scheduling Problems. Masters thesis, Universiti Putra Malaysia.

Abstract

Meeting scheduling is a distributed, tedious and time-consuming task in an organization which involves several individual in different location. The preferences and calendar availability of each individual are vary and treated as private information that unlikely to share with other individuals. Application of software agent is one of the solutions to automate this tedious task. Agent-Based Meeting Scheduling (ABMS) consists of several autonomous Secretary Agent (SA) that perform meeting scheduling task on behalf of their respective user through negotiation among them. Searching strategy is the negotiation technique that performed by SA in searching a suitable meeting timeslot. This study is interested in investigating the efficiency of searching strategy in term of communication cost, optimality of solution found and proposal successful rate during negotiation. Preliminary study of searching strategy use relaxation process to allow agents negotiate by relaxes their preference when conflicts arise. This strategy was extended with “preference estimation” technique to optimize the user preference level of negotiation outcome. However, this will increase the cost of searching process. As the result, an improvement of relaxation searching strategy by adapting artificial neural network (ANN) learning mechanism into SA is proposed in this study. ANN is used in this study because of its popularity in predicting. Unfortunately, ANN has never been used to improve the searching strategy in meeting scheduling. The back-propagation neural network (BPNN) is applied in this research to intelligently predict of participants’ preferences and guide the host in selecting proposals that are more likely to get accepted by participants. Hence, increase the accuracy of negotiation outcome and reduce the communication cost. A computer simulation is conducted to compare the proposed searching strategy with the two existing strategies namely “relaxation”, and “relaxation with preference estimation”. It is carried out by performing scheduling tasks on a set of meeting in difference calendar density. Some measurement such as, the average preference level for committed meeting, optimality of the solution, the communication cost, and rate of successful proposals are defined to evaluate the performance of these three strategies. Finally, the result of the simulation shows the ability of proposed searching strategy to find the timeslot that close to optimal solution and achieves higher average preference level. Besides, proposed searching strategy requires less communication cost to achieve optimal solution. In conclusion, the use of ANN in relaxation searching strategy successfully improves the performance of timeslot searching process in ABMS. In future works, the existing system may be extended to deal with more complex and dynamic scheduling situation such as synchronize scheduling, meeting rescheduling and user preference elicitation technique.


Download File

[img]
Preview
PDF
FSKTM_2007_19.pdf

Download (176kB)

Additional Metadata

Item Type: Thesis (Masters)
Subject: Multiagent systems
Subject: Back propagation (Artificial intelligence)
Subject: Neural networks (Computer science)
Call Number: FSKTM 2007 19
Chairman Supervisor: Associate Professor Md. Nasir Sulaiman, PhD
Divisions: Faculty of Computer Science and Information Technology
Depositing User: Rosmieza Mat Jusoh
Date Deposited: 07 Apr 2010 03:01
Last Modified: 22 May 2015 03:21
URI: http://psasir.upm.edu.my/id/eprint/5218
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item