一種加強網路服務描述語言配對的管線調合方法

Abstract

隨著網路服務的數目急遽地增加，如何有效率地從眾多的網路服務中找出符合 使用者需求的服務變得很重要。 在純文本服務匹配中，網路服務描述語言被視為 純文本，會從中提取關鍵字來表示服務， 然後通過預先訓練的Word2Vec模型轉 換成文字向量。 然而，在Word2Vec的模型中存在一個問題，因為文字向量是用 文字的上下文來訓練，這會導致兩個意義不同的文字有相似的表示方式。 此研究中，使用了管線調和方法讓文字向量和文字關係組合來改善文字向量， 而此方法是會賦予每個管線Retrofitting和Counter-fitting，並檢查這些管線的內容 和組合順序之後，才是最後使用的方法。 我們的方法在OWLS-TC V4的表現為MAP=0.9307，跟先前的研究比起來這是 最好的結果。

Recently, as the number of web services has been increasing tremendously, it becomes essential to find a web service from numerous service providers to meet users’ needs in a more effective way. In plain text service matchmaking, WSDL is treated as a plain text, keywords are extracted from WSDL, used as the service representation, and then converted into vector by means of pre-trained Word2Vec model. However, there is a main problem with Word2Vec model, that is, the pre- trained word vectors are based upon the context of words which will result in similar representations for two very different words. In this work, Word2Vec word vectors are improved by combining word relation information by means of a pipeline fitting approach to assigning to each pipe the Retrofitting and Counter-fitting with a density-based definition of neighbors and to combining these pipes into a pipeline by examing the content and the order of each pipe before adding to the pipeline. The performance of our approach to the benchmark OWLS-TC V4 is MAP=0.9307, which is the best performance comparing with previous research works.