The Knowledge Graph Refinement Of Non-functional Requirements Based On Embedded Model

Against the background of the big data,the knowledge of software non-functional requirements is distributed in large-scale,multi-source,heterogeneous,multi-modal and ever growing data sources.With the intention of combining information that is related to non-functional requirement and establishing the relationship,thus turning it into applicable knowledge,we will establish a knowledge graph of non-functional requirements.As for the knowledge graph,it tends to face the missing knowledge and errors.And when it comes to the knowledge refinement,it mainly includes knowledge completion and knowledge error detection,to be more specific,it is to solve the problem of missing knowledge and errors based on the given graph.Knowledge refinement lays a complete and correct knowledge foundation for demand engineering intelligent service.The knowledge graph embedding serves as a general method of knowledge refinement,the core of which is to embed the entities and relationships of knowledge graph into the low-dimensional continuous vector space.During this process,besides retaining the structure of knowledge graph itself,it is complemented and detected through learning the vectorized representation of entities and relationships.In this paper,we build a knowledge graph of nonfunctional requirements.In order to improve the efficiency of non-functional requirements reuse,we extend the knowledge graph of non-functional requirements with pattern and study its hierarchical structure.in this paper,a new method of knowledge graph refinement is put forward.First,In order to reuse knowledge of nonfunctional requirements when encountering new problems,we introduce a pattern approach,later,the triple is expanded,and pattern information is added,thus constructing the non-functional requirement knowledge graph based on the pattern.Then,we will further introduce pattern information into the traditional knowledge graph embedding model,and propose a pattern based non-functional requirement knowledge graph embedding model.In this model,entities and relationships are projected into the pattern space,then we will learn the vector of entities,relationships and patterns,thus deriving the quadruple similarity scoringfunction of knowledge graph.Meanwhile,hierarchical constraints are defined for the general and hierarchical structure in the knowledge graph of non-functional requirements.The loss function of the traditional model is mainly composed of scoring function and margin.In this paper,we use the hierarchy-constrained margin to replace the traditional margin.After constructing a new loss function,we use the stochastic gradient descent method to optimize the vector representation of entities,relationships and patterns.Finally,we complete the semantic calculation of knowledge through the vector representation of entities,relationships and patterns,and finish the completion and error detection of the knowledge graph of non-functional requirements.After carrying experiments including entity prediction,relation prediction,pattern prediction and quadruple classification to compare the proposed model with the traditional model,it is demonstrated that our model achieves good evaluation results in dealing with the completion and error detection of the knowledge graph of non-functional requirements.