整合多種字詞相似度的方法

Abstract

近年來，網路服務的數量與日俱增，將網路服務組合起來的研究也隨之變得重要，而在組合服務的過程中，服務的配對扮演著不可或缺的角色。為了要找出最適合並能夠滿足需求的服務，在服務文件中的資訊必須被完整地取出，並將之放入良好的結構之中藉此以量化兩個服務之間的差異，以此來改善服務配對的結果。而為了要將服務的差異量化，文字之間的語意關係必須被納入考量。我們使用了一些機器學習的演算法並將其訓練在大量的資料上以期盼能夠得到更精準的文字的語意以及字與字之間的關係。在這篇論文中，我們提出了一個整合不同語意關係的架構，此架構是用來找出服務之間的特色並使用語意關係來幫助配對服務。

In the recent years, as the number of web services has risen up, a mechanism to compose services become more important. In the process of composing services, service matching plays an important role and is indispensable. In order to find a most suitable service to satisfy the requirements, the information of the service must be extracted completely from its document, and putting the information in a structure which describes them well is also essential. By converting information into some kind of structure, the difference between two services could be quantified, and the comparison results would be helpful for matching services. To quantify the difference, word semantics must be taken into consideration. We use some machine learning algorithms and hope that using them training on huge data set could help capture more precise word semantics and relations between words. In this thesis, we propose a framework to aggregate different semantic measures for data extracted from WSDL. This framework is designed to identify the features of a service, and aggregate several measures to compare two services with their word semantics and structure.