基于词性和改进Stacking模型的需求依赖关系提取  

Requirement dependency extraction based on part of speech and improved Stacking model

作　　者：关慧 许航[1] 蔡丽娥 GUAN Hui;XU Hang;CAI Li-e(Department of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;Key Laboratory of Industrial Intelligence Technology on Chemical Process,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学计算机科学与技术学院,辽宁沈阳110142 [2]沈阳化工大学辽宁省化工过程工业智能化技术重点实验室,辽宁沈阳110142

出　　处：《计算机工程与设计》2024年第11期3345-3351,共7页Computer Engineering and Design

基　　金：辽宁省教育厅2021年度科学研究经费基金项目(LJKZ0434)。

摘　　要：为解决需求工程中人工分析需求依赖关系面临的成本和效率问题,提出一种基于词性特征和改进Stacking集成学习模型(P-Stacking)的需求依赖关系提取方法。在词性权重确定过程中,提取出能表征需求句主干语义的主谓宾三元组作为3种词性,使用粒子群算法迭代计算出各词性权重,以此改进TF-IDF。P-Stacking使用相关性较小算法为基模型选择相异分类器,使用网格搜索算法匹配最优基分类器组合。实验结果表明,在3个数据集的评估测试中,分别引入词性特征和集成学习模型后,需求依赖类型预测准确性有了显著提升。To address the cost and efficiency issues of manually analyzing requirement dependencies in requirements engineering,a requirement dependency extraction method based on part of speech features and improved Stacking ensemble learning model(P-Stacking)was proposed.In the process of determining the weight of part of speech,subject predicate object triplets that represented the main semantics of the requirement sentence were extracted as three types of parts of speech,and the particle swarm optimization algorithm was used to iteratively calculate the weight of each part of speech,thereby improving TF-IDF.In the P-Stacking ensemble learning model,the algorithm with less correlation was used to select different classifiers for the base model,and a grid search algorithm was used to match the optimal combination of base classifiers.Experimental results show that the accuracy of requirement dependency type prediction is significantly improved after the introduction of part of speech features and ensemble learning model respectively in the evaluation test of the three data sets.

关 键 词：需求依赖 依赖提取 词性特征 粒子群优化算法 集成学习 相关性较小算法 网格搜索算法 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]