National and Kapodistrian University of Athens
E-negotiations for trading commodities and services: predictive strategies
Abstract
dc:descriptionThis thesis takes into account the advances in the field of electronic bi-lateral negotiations, adopting state-of-the-art protocols and strategies that characterize the behavior of each party. The research objective is the application of strategies that are based on the estimation of the counterpart’s next offer, and give the predictive agent the advantage to establish agreements that are more beneficial. An issue that is contemplated is that of risk when employing a predictive strategy. A new strategy that permits the adoption of different risk attitudes is proposed. This thesis also focuses on the AI-based models used for prediction. The main problem of related applications is their inability to capture the dynamics of turbulent negotiation environments, and provide accurate estimations also in cases where data distributions change. For this reason utilization of models that are based on data acquired from the current discourse and adapt their structure in time are examined. More specifically application of neural networks with static structures (SSLAs), and application of neural networks with adaptive structures (ASLAs) are developed and assessed. To illustrate the superiority of SSLAs compared to agents that employ pre-trained networks (PTAs) in cases where data distributions change, we have conducted a number of experiments considering single-issued negotiations and we have computed the absolute error yielded in each decision making step. SSLAs are proved more accurate, as the mean of the errors is reduced by 92.67% compared to the PTAs. As far as adaptive structures are compared, we have tested neural networks that optimize their network on the basis of a genetic algorithm. ASLAs are proved more accurate as the average mean of the absolute errors is reduced by 38.34%, the average of the maximum values is reduced by 44.75% and the average of the standard deviation is reduced by 38.03%. ASLAs are smoother predictive models as they result to decreased standard deviation and maximum error values. However they have high storage and computational requirements, which makes them difficult to apply.Static session-long learning agents were extended to support multi-issued negotiations and the architecture of the network was empirically searched. The network yielding the lowest error incurred to the predictive agent 10.78% average absolute increase of his individual utility (gain).
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Masvoula, Marisa
- Μάσβουλα, Μαρίζα
Subjects
dc:subject × 14- Ηλεκτρονικές διαπραγματεύσεις
- Πράκτορες διαπραγμάτευσης
- Νευρωνικά δίκτυα
- Γενετικοί αλγόριθμοι
- Στρατηγικές πρόβλεψης
- Electronic negotiations
- Negotiating agents
- Neural networks
- Genetic algorithms
- Predictive strategies
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/33261