The Research And Implementation Of Key Technologies In Electronic Commerce Automated Negotiation

Because of their capabilities to imitate the social intelligence of human being flexibly and adaptively, Multi-Agent Systems (MAS) can be widely applied to construct open and dynamic software systems. And it is one of the major research areas in artificial intelligence. With the improvement of people's requirement in the electronic commerce intelligence and personalization, the implementation of intelligence and personalization of electronic commerce has become the direction of next generation of electronic commerce. Besides, the research of electronic commerce automated negotiation has become one of the typical applications of MAS. In this paper, we take the negotiation protocol, negotiation strategy and agent learning as the subjects. The major contributions of this thesis are as follows:1. To analyse the properties a good negotiation protocol, negotiation strategy and negotiation mechanism should have. The difference and relationship between the negotiation protocol and the negotiation process is illustrated. And the FIPA acts and the format of FIPA Agent Communication Language (ACL) information are extended.2. To propose an extended Colored Petri Net (ECPN). The ECPN can represent the basic concepts in negotiation protocol, as well as the semantic relationships among these concepts. This thesis also presents the matrix analysis method and the extended Probabilistic State Graph analysis method and applies them in the analysis of the consistency and reachability of auction protocol ECPN model and 1:1 ECPN model.3. To propose a protocol ontology framework to realize the common understanding and sharing of protocols among agents. This thesis gives out an algorithm whose function is to translate the ECPN model into OWL ontology. After being imported into Protege, the terminology consistency, subsumption consistency and reference consistency of this ontology are checked using the test function offered by the Protege and the RACER tool. An API interface program is created to facilitate agent accessing the protocol ontology.4. To define the decision models used in negotiation process, analyze the existence of the equilibrium in agents' attitude selection, propose an attitude-adaptation negotiation strategy. Moreover, an attitude capture algorithm is given out based on the historical interaction information. A strategy analysis simulation system is also realized. The attitude-adaptation strategy and the fixed attitude strategies are analyzed and compared using this system.5. To present the agent's behavior selection tendency matrix, the agent's behavior selection algorithm and the history-based behavior learning algorithm (HQ-learning algorithm). Besides, the convergence and converge speed of the HQ-learning algorithm is explained.6. To introduce details of the framework of the electronic commerce automated negotiation system prototype. The execution process of this prototype is described using an example.