A Requirements Traceability Approach Based on Neural Network Ensemble Learning and Semantic-Enhanced Features
The existing machine learning-based requirements traceability approaches suffer from instability.This paper proposes an Ensemble Learning Trace Approach(EMTrace)based on neural network ensemble learning and semantic-enhanced features,to alleviate these issues.The approach transforms the requirements traceability problem into a classification task by leveraging multiple machine learning classifiers for prediction.These classifiers are integrated,and weights are assigned to their predictions to establish trace links.To automatically obtain the weights of each base model,this paper constructs a neural network-based meta-learner and trains it using the predictions of each base model.Additionally,to accurately represent the trace links between artifacts,the EMTrace approach uses multiple word embedding and sentence embedding models to extract the semantic information of software artifacts and enhance the semantic representation of trace link features.Experimental results demonstrate that the EMTrace approach effectively improves the stability and performance of requirements traceability,achieving a 0.162 improvement in F1 compared to the optimal baseline method.
NETL
NSTL
万方数据
