Detecting similarities and conflicts in software requirements

需求是用自然语言编写的，因此容易产生歧义、模糊和主观性。它们并不总是清晰和连贯的。它们也可能包含许多重复和/或相互冲突的信息。自动化解决方案使用信息检索和机器学习技术来捕获这些异常；然而，当前的技术忽略了软件工件的语义，或者忽略了将领域知识集成到这个过程中，因此倾向于交付不精确和不准确的结果。在这个研究项目中，我们想要提出一个使用深度学习的解决方案，将需求工件语义和领域知识纳入考虑。我们的目标是尽可能准确和及时地捕获冲突和重复的需求。我们提出了一种利用词嵌入和卷积神经网络（CNN）模型生成链接的网络架构。词嵌入学习表示领域语料库知识的词向量，CNN使用这些词向量来学习需求工件的句子语义。我们将使用向量空间模型和潜在语义索引来构建基线方法，以比较并遵守我们提出的模型与最先进的模型以及沃森发现。我们将使用DOORS下一代需求的数据集。我们建议的模型可以进一步扩展，以构建设计需求和源代码之间的链接，从而实现完整的可追溯性矩阵。

Requirements are written in natural language and hence are subject to ambiguity, vagueness, and subjectivity. They are not always clear and coherent. They may also contain many duplicates and/or conflicting information. Automated solutions use information retrieval and machine learning techniques to capture such anomalies; however, current techniques ignore the semantics of the software artifacts or ignore integrating domain knowledge into this process and therefore tend to deliver imprecise and inaccurate results. In this research project, we would like to propose a solution that uses deep learning to incorporate requirements artifact semantics and domain knowledge into consideration. Our aim is to capture conflicting and duplicate requirements as accurately and timely as possible. We propose a network architecture that utilizes Word Embedding and Convolutional Neural Network (CNN) models to generate links. Word embedding learns word vectors that represent knowledge of the domain corpus and CNN uses these word vectors to learn the sentence semantics of requirements artifacts. We will build baseline methods using the Vector Space Model and Latent Semantic Indexing to compare and comply our proposed model with the state-of-the-art models as well as Watson Discovery. We will use datasets from DOORS Next Generation requirements. Our proposed model could further be extended to build the links between design requirements and source code to enable having a full traceability matrix.