基于深度学习的需求自动抽取研究

Research on automatic demand extraction based on deep learning

需求分析是需求工程研究的基础,需求条目在需求工程中起到了至关重要的作用,由于目前市面上比较缺乏有关于自动抽取需求的工具以及研究,研究首先通过多种方法收集需求语料,再基于收集到的语料构建一个需求抽取数据库,提出一种基于深度学习的RoBERTa⁃wwm⁃BiLSTM算法开展需求自动抽取的研究。经过对比验证得出RoBERTa⁃wwm⁃BiLSTM算法需求抽取效果与其他算法相比在各方面均有较大提升,准确率、召回率和F1值分别为94.74%、93.33%和93.65%,结果证明了该方法在需求抽取任务中的有效性。

Requirements analysis is the basis of requirements engineering research,and requirements entries play a crucial role in requirements engineering.Since there is a lack of tools and research on automatic requirements extraction in the market,the study first collects requirements corpus through various methods,then constructs a requirements extraction database based on the collected corpus,and proposes a deep learning‑based RoBERTa‑wwm‑BiLSTM algorithm based on deep learning is proposed to carry out the study of automatic demand extraction.After comparing and validating the results,we found that the RoBERTa‑wwm‑BiLSTM algorithm has improved in all aspects compared with other algorithms,and the accuracy,recall and F1 values are 94.74%,93.33%and 93.65%,respectively,which proves the effectiveness of the method in the demand extraction task.