利用信賴與協同預測 QoS 方法來幫助開發 Service-Based Systems

Abstract

Service-Based Systems 的開發通常透過現存的 Services 快速有效地去組成一個新的 應用,至於現存 Services 的挑選一般採用 QoS 的預測值來進行,現今已有許多預測 QoS 的方法被提出來,這些方法在歷史資訊不足時,大多都無法提供準確的 QoS 預測值,因 此,在這種情況下也無法對 Service 的選用提供有效的幫助。本文中,我們提出一個信 賴推理的方法來獲取使用者之間潛在的相似度,並以此去改善在資訊不足時的預測準確 度,實驗數據顯示出我們的方法無論在資訊不足時或一般情況準確度都優於目前的方法。 另一方面,Workflow 一種常用的 SBS 架構,運用 Structured Workflow 組合已知 Service 來產生一個新的(應用)服務是一種相當普遍的方法,對於 Service Application 的製造 來說,當每個被採用的 Service 的預測值都得到時,我們也提出一套運用這一製作技術 來產生的新 Application 之預測值的方法。

An application (service) can be composited with existing services fast and effectively. Generally, the appropriate services can be selected according to the predicted QoS values. There were several QoS prediction approaches proposed. However, when the invocation records are small, the QoS values derived from these approaches are inaccurate, i.e., the first present. QoS values cannot provide effective help to select services. In this thesis, we proposed a trust-aware QoS prediction method by reasoning the indirect similarity between users to improve the accuracy of QoS values. The experiment results show that the accuracy of QoS values of our method is better than others when invocation records are small. On the other hand, workflow is a common SBS framework. Compositing a new application (service) with existing services by a structured workflow is a popular method. For evaluating a service application, the thesis also present a method to derive the QoS of the application to help its design/construction.