一個基於動態資料驅動應用系統之半分散式信譽機制

Abstract

信賴關係在許多未知網路中是相當重要的研究議題之一，透過有效率且準確的信賴值判斷，能夠分辨哪些節點是值得或不值得信賴之關係。網路中的信賴動態本質 (Dynamics of Trust)，會造成節點中毒、偽裝等情況，從而產生異常行為。為了了解信賴動態本質，本研究在半分散式架構下，利用動態資料驅動應用系統 (Dynamic Data-Driven Application System)，提出一個信譽機制包含區域性信譽 (Local Reputation)與全域性信譽 (Global Reputation)兩種指標。節點透過自身經驗與鄰居推薦計算出區域性信譽，可以初步判斷是否交易。接著低於特定門檻值之區域性信譽會被上傳至中央控制器，進而驅動去計算出全域性信譽，可以更準確地判斷是否交易，我們利用類神經網路來實作。實驗結果顯示，平均只上傳52.21%的區域性信譽就能計算出全域性信譽，而此全域性信譽能夠在短時間內平均上升或下降26.5%，這個現象表示所提出的機制能夠有效地抓取信賴動態本質。

Trust is one of the important issues related to unknown networks. A mechanism which can distinguish a trustworthy node from an untrustworthy one is essential. The effectiveness of the mechanism depends on the accuracy of node’s reputation. Dynamics of Trust often happens in a trusted network. It causes intoxication and disguise for nodes, resulting in abnormal behaviors. This thesis proposes a semi-distributed reputation mechanism based on Dynamic Data-Driven Application System. This mechanism includes two reputations: Local Reputation (LRep) and Global Reputation (GRep). Nodes use their own experience and neighbors’ recommendations to compute LRep, which is then used to determine whether to continue trading or not. LReps are uploaded to the central controller. The central controller computes GRep, which can then be used to determine whether to continue trading. Neural Network is used in the experiments. The experimental results show that a GRep can be computed with only on average 52.21% LReps uploaded. Also, GRep rises or falls on average 26.5% in a short period of time. This phenomenon demonstrates the proposed mechanism can effectively handle Dynamics of Trust.