A Deep Learning and Ontology Based Framework for Textual Requirements Analysis and Conceptual Model Generation

Recent years have witnessed significant successes in system development with the help of system engineering. Analyzing requirements correctly and creating subsequent conceptual models effectively is a key part of the system engineering process, which in turn determines the quality of target systems. However, most of the current strategies mainly based on human efforts, in which textual requirements are processed and then transferred to models leading to disadvantages including high time consumption and human errors. To address the problem, we proposed a framework to process textual requirements and generate models automatically by incorporating both natural language processing and ontology-based techniques. Using a deep convolutional neural network, textual requirements are analyzed and extracted into raw elements, e.g., entities and relationships, which further motivates the fine-grained processing. By introducing domain-specific ontology, the consistency and completion of the extracted information are verified and checked. Extensive experiments on SemEval-2010 tasks and datasets demonstrate the effectiveness of our proposed methods.