Research On One-to-Many Negotiation In Electronic Commerce

Nowadays, Electronic Commerce has already come into our homely lives. People are willing to go to shopping at some shopping websites, such as taobao (www.taobao.com), and slowly form a habit. On July 15, 2008, Data Center of the China Internet (DCCI) published the research data for the first half of 2008, which shows that the Internet consumption of the Chinese Internet users was 256 billion, increased by 58.2% as compared with the same period last year. But the typical pricing methods of these shopping websites are the fixed price, auction and bid by agent, which are all semi-intelligent. Fixed price is too simple; Auction needs real time interaction between men and machine; Bid by agent is relative simple and has a large risk. Furthermore, these methods are all unilateral, while negotiation mechanism is most sapiential, flexible and bilateral.With the development and increasingly ripeness of Agent technology of Distributed Artificial Intelligence (DAI), it is possible to make Electronic Commerce automatic and intelligent. Electronic Commerce based on agent enhances the automatization and intelligence of business process, which will improve the development of Electronic Commerce. Negotiation is a most important phase of business process, directly affects sequent execution of whole business process. Automated negotiation is a key form of interaction in agent-based systems. Now negotiation is a hot research of Multiple Agent System (MAS). In terms of the number of agents participating in negotiation, agent-based automated negotiation can be divided into three cases: one-to-one negotiation (bilateral negotiation), one-to-many negotiation and many-to-many negotiation. One-to-one negotiation is researched earliest of these three cases; one-to-many negotiation can be seen as multiple concurrent one-to-one (bilateral) negotiations; many-to-many negotiation can be extended from multiple one-to-many negotiation.This paper researches on one-to-many negotiation in Electronic Commerce. The following are the main research contents and the innovations.①Research on one-to-many negotiation modelBased on research of the existing general one-to-many negotiation model, a more flexible one-to-many negotiation model is proposed. This model supports continuous, open and dynamic negotiation. Compared with the general model, the advantages of our flexible model are: 1) Our model supports continuous negotiation which can decrease negotiation time, improve negotiation efficiency. Buyer agent should not wait until having received offers from all its trading partners before generating counteroffers, that is, negotiation threads need not wait each other. 2) Our model satisfies the requirement of open and dynamic settings. In the middle of negotiation, new seller agent can join, and the existing seller can leave at any time. 3) Our model uses the coordination strategy based on relative utility to select the best agreement which has the higher utility and lower cost.②Research on concession negotiation strategiesIn order to enhance and optimize the negotiation result reached by concession negotiation strategies (Time-based negotiation strategy, opponent-behavior-based negotiation strategy), equal-utility exchange negotiation strategy is proposed based on the concession negotiation strategies. Equal-utility exchange makes use of the relationship between the agent's negotiation-related issues of the multi-issues utility evaluation mechanism. In order to get a better negotiation result, values of some issues are adjusted, ensuring the utility of agent not lower. Equal-utility exchange negotiation strategy can enhance joint utility partly and reach a better negotiation result than concession negotiation strategies.③Research on learning negotiation strategiesImporting machine learning to agent can make agent have the ability of learning, reasoning and intelligence. This paper firstly researches these popular machine learning, such as Bayesian learning, reinforcement learning and genetic algorithm. Then this paper proposes an optimized negotiation strategy based on reinforcement learning. In the middle of negotiation process, it makes the best use of the opponent's negotiation history, in order to quicken the negotiation result convergence and enhance the negotiation result quality.④Research on coordination strategiesOne-to-many negotiation can be look as multiple, concurrent one-to-one (bilateral) negotiations. This needs a coordinator use coordination strategies to coordinate these multiple, concurrent one-to-one negotiations in one-to-many negotiation. Firstly,this paper researches several existing coordination strategies: Desperate Strategy, Patient Strategy, Optimized Patient Strategy, Strategy Manipulation Strategies, Fixed-Waiting-Time-Based Strategy, and Fixed-Waiting-Ratio-Based Strategy. Then relative utility theory is proposed. Next, coordination strategy based on relative utility is proposed. This coordination strategy can deal well when multiple, concurrent one-to-one negotiations all satisfy utility evaluation, especially having multiple maximum same utility offers at the same time. Coordination strategy insures multiple, concurrent one-to-one negotiations get the best negotiation result which has high utility and low cost.⑤Research on management mechanism for the best seller withdrawingFirstly, influence of the current best seller withdrawing is analyzed based on the existing one-to-many negotiation model, including decrease of final utility value and prolonging of average negotiation time; and three coordinating strategies were proposed. Then, one-to-many negotiation model with commitment management which used to restrict the current best seller withdrawing was created. Commitment management could reduce the probability of withdrawing of the current best seller, decrease influence of system performance and provide a more flexible mechanism for one-to-many negotiation model.This paper thoroughly researches several main aspects of one-to-many negotiation, including negotiation model, negotiation strategy, coordination strategy, and management mechanism. Our researches not only succeed to the research results of predecessor, but also develop the research and get innovations, which are frontier , and of theoretical and practical value.