Time Series Predictive Models for Opponent Behavior Modeling in Bilateral Negotiations

Abstract

In agent-based negotiations, understanding and predicting an opponent’s bidding patterns is crucial for strategic decision-making. Foreseeing the utility of an opponent’s upcoming offer provides valuable insight for an agent to determine its next move. This paper focuses on predicting an opponent’s future offers using two deep learning approaches: Long Short-Term Memory Networks and Transformers. The learning process has three main targets: 1. Estimating the agent’s utility of the opponent’s coming offer. 2. Estimating the agent’s utility of that offer without using opponent-related variables. 3. Estimating the opponent’s utility of that offer by using opponent-related variables. The work evaluates these models in various negotiation scenarios, and the results show promising prediction performance.