Research On The Construction Of Bilingual Knowledge Graph And Entity Linking Of Software Nonfunctional Requirements

In the process of software development,nevertheless,requirements engineering is facing with many problems during the process,the most prominent ones are incomplete requirements,implicit requirements,difficulty in communication between users and requirements engineers,inevitable frequent changes of requirements,etc.The non-functional requirements in requirements engineering is an important indicator of software quality,which also has the problem of ambiguity,inconsistency,incompleteness and frequent changes.In such a complex,dynamic and diverse situation,it is necessary to integrate non-functional requirements knowledge into computer systems to complete knowledge engineering,which can provide stakeholders with comprehensive non-functional requirements knowledge.The provision and use of non-functional requirements knowledge are greatly hindered since non-functional requirements are underestimated,the knowledge is scattered,and the non-functional requirements knowledge between different languages is unbalanced.In order to solve the above non-functional requirements,building a bilingual knowledge graph of non-functional requirements knowledge is conducive to knowledge sharing and reuse.This article obtains the attributes of non-functional requirements entities in Wikipedia and Baidu Encyclopedia,uses the protégé tool to model the acquired data at the pattern layer,and uses the pattern layer to detect the consistency and satisfiability of non-functional requirements knowledge in the data layer reasoning.The modeled data is embedded and expressed,and the knowledge of bilingual non-functional requirements in Chinese and English is mapped into a shared space using Bi-LSTM,Bi-GRU,and Self-Attention mechanisms.Finally,the similarity calculation is performed on the bilingual non-functional requirements entities in the shared space through the cosine similarity and the entity links are performed.Through experiments,compared with the similarity calculation after embedded representation using the single traditional methods Bi-LSTM + Attention and Bi-GRU + Attention,the method used in this paper improves the accuracy of non-functional requirements entity link accuracy than the above two methods Increased by 6.3% and 3.9%,respectively.