On the Measurement of Negotiation Dialogue Games

Nowadays, multiagent systems became a widely used technology in everyday life. More studies are needed to evaluate these systems from different aspects such as evaluating agent dialogues, the participants to these dialogues, and the protocols governing the dialogues, etc. In this paper, we define new measures for dialogue games from an external agent’s point of view. In particular, two measurement sets are proposed: in the first set, we use Shannon entropy to measure the certainty index of the dialogue. This involves i) using Shannon entropy to measure the agent’s certainty about each move during the dialogue; and ii) using Shannon entropy to measure the certainty of the agents about the whole dialogue with two different ways. The first way is by taking the average of the certainty index of all moves, and the second way is by determining all possible dialogues and applying the general formula of Shannon entropy. In the second set, we introduce two metrics: i) measuring the goodness of the agents in the real dialogue (i.e. the dialogue that effectively happened between the participants); and ii) measuring the farness of the agents from the right dialogue (i.e. the best dialogue that can be produced by two agents if they know the knowledge bases of each other). Many dialogue game types have been proposed in multiagent systems. In this paper, we focus on one specific type, namely quantitative negotiation such as bargaining.