Assisting engineers extracting requirements on components from domain documents

作     者：Lian, Xiaoli Liu, Wenchuang Zhang, Li 

作者机构：Beihang Univ Beijing 100191 Peoples R China ByteDance Inc Beijing 100083 Peoples R China 

出 版 物：《INFORMATION AND SOFTWARE TECHNOLOGY》 (信息与软件技术)

年 卷 期：2020年第118卷

页      面：106196-000页

核心收录：

学科分类：08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Science Foundation of China [61672078  61732019] 

主　　题：Requirements discovery Domain reference model Documents ranking Extractive text summarization 

摘      要：Context When entering an unfamiliar domain, organizations usually have to invest significant time and effort performing domain analysis with the purpose of acquiring system requirements. This process usually involves collecting domain documents extensively, retrieving and reviewing the related ones carefully, searching for the requirements knowledge, then extracting and specifying system requirements. Furthermore, the task must often be performed repeatedly throughout the early phases of projects. Depending on the nature of the domain and the availability of documentation, this process is extremely time-consuming and may require non-trivial human effort. Objective: In order to assist engineers identifying requirements knowledge from a collect of domain documents, previously we proposed an approach MaRK in the Conference RE 16 which ranks the domain documents by their relevance to components and highlights the content that are likely to contain component-related information. Experiments showed MaRK can almost identify the top and bottom documents in the reference list. However, it tends to underestimate the relevance of the domain documents that have a number of sections with medium knowledge density. Method: We improve the ranking algorithm in MaRK and propose MaRK-H. In addition, to assist engineers locating the relevant information in lengthy documents, we preserve the highlighting work in MaRK and strengthen MaRK-II by extracting the summary of component-related text. MaRK-II is evaluated with the documents in three domains. Results: We found that MaRK-II significantly outperforms MaRK and VSM on ranking the documents by their relevance. And a user study showed that MaRK-II is indeed helpful for engineers to extract requirements on components. Conclusions: Our approach provides three mechanisms including documents ranking, pertinent content highlighting and summarizing to help engineers obtaining requirements from a collection of domain documents.