Semantic Recognition of Aerospace Mode Software Demand Based on Bi-LSTM

The process of requirements elicitation is in the initial stage of the front-end software development. As the main demand source of aerospace model software, technical documents exist in the form of natural text without rules and constraints, and contain multi-level elements and complex sofeware demand semantics, which brings great challengs to the demand acquisition and development enforceability. Therefore, a semantic recognition method of aerospace software requirements based on Long Short Term Memory (LSTM) is proposed in this study, This method extracts semantic elements from software technical documents, and then establishes a knowledgeable description of the semantic information of software requirements. The structural description of aerospace software requirements is defined and the description form of its knowledge graph is specified. Then an aerospace software requirement semantic recognition model based on Bi-LSTM was constructed, which includes requirement morpheme decomposition, requirement morpheme classification and semantic relationship construction. The model was trained with aerospace technical documents, and the results show that the proposed model can achieve more than 96% morpheme decomposition accuracy and 97% word segmentation accuracy, which proves the effectiveness of the proposed method.