A Combination of LERT and CNN-BILSTM Models for Chinese Music Named Entity Recognition
Chaoguo Wang, Liang Zhang, Wei Yan · 2024
Named Entity Recognition (NER) is an important task in the field of natural language processing(NLP), serving as the foundation for knowledge graphs, recommendation systems, machine translation, and other applications. However, progress in NER for the music domain has been relatively slow. This is mainly due to the diverse and irregular nature of music entities, as well as the labor-intensive process of entity annotation, resulting in a lack of publicly available datasets, therefore, this paper organizes music texts on the internet through techniques such as data mining and NLP, constructing a NER dataset tailored to the field of music. Pre-trained models have been widely adopted in various fields, including NLP and computer vision, as they can learn rich language knowledge and semantic representations. To address the limitations of existing NER methods in capturing contextual information and handling word sense disambiguation, this paper proposes a deep learning model called LERT-CNN-BILSTM-CRF for entity recognition in the Chinese music domain. Experimental results demonstrate that our proposed method achieves good performance, with precision reaching 93.87%, recall reaching 95.06%, and an F1 score of 94.46%. Compared to previous machine learning-based methods for music NER, our proposed approach better captures the semantic information of complex music named entities, thereby improving the performance and generalization ability of the model.