近年來隨著對於大型軟體系統需要的增加，開放式多代理人系統逐漸被重視並成為一個提供滿足此需求之系統平台，而在建立此平台之研究領域中，代理人組織之觀念被廣泛的採用並成為此平台之建構基礎。為了形成代理人組織，代理人動態角色的扮演儼然成為一個重要且關鍵的一環。然而，對於動態角色的扮演存在著以下幾個重要的問題：(一)代理人組織如何在開放的環境中尋找到代理人；(二)代理人組織如何在多個代理人中選擇適合的代理人來扮演角色；(三)代理人如何透過角色的扮演來合作以達成組織的目的；(四)當代理人欲離開組織時如何脫去所扮演的角色。在本論文中，我們提出了以下兩個方法：(一)我們提出了動態角色扮演生命週期來闡述上述之問題，並延伸簽約網路協定發展了一個結合評價模型之協定，稱之為以代理人為中心之簽約網路協定；(二)我們提出了一個風險支援評價模型來促進協定中代理人之選擇，並利用兩個實驗來驗證此模型。 As open multi-agents systems are getting increasing attentions in building large-scale software systems, the study on agent organizations -- a building block of multi-agents systems, becomes a promising research track. Roles as a first class citizen in agent organizations impose an imperious demand for dealing with the major challenge: how agents bind roles in the formation of agent organizations in which agents can dynamically enter or leave the organizations. This challenge can be further boiled down to four issues: how to discover agents distributed over various organizations, how to select reliable agents to bind roles, how to enable agents to cooperate on roles binding, and how to unbind roles as agents leave an organization. The focus of this dissertation is two-fold: (1) an extension to Contract Net protocol with reputation model to address these four issues related to the lifecycle of dynamic role binding, called Agent-centric Contract Net Protocol (ACNP), as a negotiation mechanism that agents can adopt to establish agent organizations at run time through roles binding; and 2) a risk-enabled reputation model, called REAL, to facilitate the decision making of agents selection in ACNP, which is validated through two empirical experiments.
