To decrease the incomplete information between agents in MAS(multi-agent system)based e-commerce,and to accelerate the process of negotiation,and enhance the efficiency of negotiation,a classified learning mechanism is provided in bilateral multi-issue negotiation to learn some incomplete information especially the incomplete information about agent opponent\'s negotiation issue weight.

To make the agents more efficient in bilateral multi-issue negotiation in multi-agent based e-commerce,an agent negotiation model is presented.

The experimental result show the genetic algorithm based on metropolis rule can gain the optimal negotiate result more efficiently in multi-literal multi-issue negotiation.

Classical negotiation methods based on game theory obtain the optimal resolution through thorough search of the space, which are consequently not suitable for multi-issue negotiation.