| Artificial intelligence research involves many subjects, such as cybernetics, information theory, linguistics, mathematical logic and philosophy. Agent are expected to be able to think and reason, like human beings, and considered an entity to exist in a certain environment, which could sense the environment, receive messages from the environment, then react to the environment. The research mainly covers knowledge representation, automated reasoning and search methods, machine learning and knowledge acquisition, knowledge processing, natural language understanding, etc. Therefore, knowledge representation and reasoning become an important part of intelligent systems, which determines the agent’s reasoning and ability to learn.Negotiation theory played a vital role in the process of multi-agent system research. From the late1980s to the1990s, negotiation models of multi-agent system research are largely static, which apply game theory to do under the complete knowledge and in a closed environment. It is noteworthy that the game theory plays a very important role in the negotiation of multi-agent systems, because the game theory is widely considered the best mathematical tool to study the human social interaction, which can evaluate whether a negotiation is effective. With the development of artificial intelligence, complete knowledge and negotiations under the closed state could not meet real requirements. It is hoped that the agent can complete the reasoning process of uncertainties like human beings. It is a non-monotonic process. In recent years, the researchers adopt the Answer Set Program (ASP) with non-monotonic properties to represent the background knowledge of the agent who participated in the negotiation. There were a series of negotiation mechanisms and methods, which are studied by the means of logical reasoning. ASP has a strong expressive ability in the knowledge representation. This paper continues to follow this practice. ASP is a form of declarative programming. Based on knowledge base and expected outcome, agent completes the calculation automatically and feedbacks the path to achieved the outcome to user, which reflects the automatic reasoning ability of artificial intelligence. Taking ASP to express the background of agents, an answer set is one of the agent’s negotiation demands. An answer set is a set of literals, so a negotiation demand is actually a set of literals. In this paper, according to its own preferences, each agent sorts the literals which will be accepted or given up. The literal with higher priority is able to exchange the literal with lower priority, but can’t vice versa.Negotiation is the process of resolving inconsistencies or reducing uncertainties to reach a consensus, the whole process is actually a game process, and each participant exchanges some resources to reach an agreement. At present, the mostly research of negotiation is between two agents, but this kind of negotiation has many limitations, it often leads to the "philosopher dining" problem due to lack of resources, and results the negotiation to fail. If there was a third-party, each participant in the negotiation has more choices on objects and contents. Thus it is possible to avoid the above problems. This paper builds the multi-agent model with multi-round based on the two agents’ multi-round. This paper proposes the notion of profit-compensatory and the way of belief revision under the multi-agent negotiation. If there are some factors hinder the negotiation further, agent will check whether it can get profit-compensatory in the next round. If it can get it, the agent will delete these factors and revise belief to continue the negotiation. At the same time, this paper proves the negotiation of multi-agent with multi-round is effective by the game theory. This paper practices the model in the given instance with ASP in computer, the model is feasible. The experiment result is the same as our theoretical predictions, and it satisfies the sub-game perfect Nash equilibrium. |
PDF Full Text Request