Home // COGNITIVE 2013, The Fifth International Conference on Advanced Cognitive Technologies and Applications // View article

Forecasting Negotiation Counterpart's Offers: A Focus on Session-long Learning Agents

Authors:
Marisa Masvoula

Keywords: Predictive negotiator, genetic algorithm, adaptive negotiation strategy, neural network applications

Abstract:
Predictive decision making is characteristic to current state of the art socio-technical systems that guide negotiation processes under electronic settings. Back end participants are particularly benefited by the use of models of computational intelligence, which help them adapt their strategy and evaluate risks and dynamics of the current negotiation. In this paper, the skill of forecasting the counterpart’s future offers with the use of neural networks is investigated. Current systems base their learning models on data acquired from previous interactions. Such systems are once trained in an offline mode and are thereafter expected to operate in a real environment. However, when data distributions change, the systems no longer provide accurate estimations. A new perspective to the issue is introduced, by highlighting the need of learning during the negotiation session, with the use of “session-long learning” agents. These agents prove capable of capturing the negotiation dynamics by training their learning models with the data from the current negotiation thread. In this paper a static session-long learning agent, based on a simple neural network model, as well as an adaptive session-long learning agent, based on a neural network which evolves its structure and input features with the use of a genetic algorithm in each negotiation round, are presented and assessed.

Pages: 71 to 76

Copyright: Copyright (c) IARIA, 2013

Publication date: May 27, 2013

Published in: conference

ISSN: 2308-4197

ISBN: 978-1-61208-273-8

Location: Valencia, Spain

Dates: from May 27, 2013 to June 1, 2013