Named Entity Recognition on User Requirement Analysis with BERT-CNN

In the realm of software development, the emergence of new requirements that have already been addressed or that bear functional similarities to existing modules poses significant challenges. These include duplicate efforts, inconsistencies, delays, and increased costs. To mitigate these issues, this research proposes the management of requirement documents using Named Entity Recognition (NER) to generate an ontology that serves as a structured representation of information. Traditionally, the task NER in user requirements has been addressed using machine learning and deep learning techniques. However, the advent of transformer-based technologies, notably Bidirectional Encoder Representations from Transformers (BERT), presents a novel opportunity for enhancement. This study addresses this research gap by integrating BERT with Convolutional Neural Networks (CNN) to improve the performance of NER tasks. By leveraging the contextual understanding capabilities of BERT and the spatial feature extraction strengths of CNN, our approach aims to achieve superior accuracy in recognizing entities within the user requirements. The combination of these advanced techniques is expected to provide a more precise and reliable extraction of entities. The results demonstrate that the overall F1 score across all entity categories improved from 77% with BERT to 87% with BERT-CNN, indicating that integrating CNN with BERT improves the performance of the model in NER tasks.