基於規則的服務描述擷取

RULE-BASED EXTRACTION FOR SERVICE DESCRIPTIONS

現有的自動化服務組合方法大多都只接受tuple-based格式輸入，不過對於人類而言自然語言才是最方便直覺的表達方式，因此本研究開發從自然語言語句中自動提取Input和Output元素的方法。方法利用CoreNLP自然語言處理工具獲取語句的語法和片語結構資訊，接著根據規則對語句逐步匹配，最終輸出目標元素。經過實驗分析，方法在User Story形式的句子中擁有較佳且穩定的表現，但在撰寫風格較為多樣化的一般軟體需求則表現較差，對此本研究歸納造成錯誤的原因，供後續相關研究借鏡與改進。

Although natural language is the most convenient method of expression for humans, most existing automated service composition methods only accept the tuple-based form as input. As a solution, we have developed an automatic method to extract elements, such as input and output, from natural language sentences. The method utilizes the CoreNLP natural language processing tool to obtain information about the syntactic and phrase structure of the sentence. Then, the statement is systematically matched according to the rules, and finally, the target element is outputted. After evaluation, the method demonstrated improved and consistent performance in user story sentences, but its ability to handle diverse writing styles in general (Free-form) software requirements fell short of expectations. We identify the reasons for this deficiency and provide these reasons for reference in related research.