Research on Methods for Constructing Named Entity Models in the Ceramic Field

Named Entity Recognition (NER) is a crucial task in Natural Language Processing (NLP). While NER has been applied in many fields, the application in the ceramic field is limited due to the scarcity of domain-specific datasets. This study employs BIO annotation to label a ceramic dataset and conducts a comparative analysis across multiple models. The experimental results demonstrate that the BERT+FUSIONATT+CRF model performs the best on the ceramic dataset, outperforming other models with an accuracy of 90.47%, showcasing the model's superior performance in NER tasks. The BERT+FUSIONATT+CRF model combines BERT’s deep pre-trained language model with dynamic attention mechanisms, effectively capturing complex semantic information and dynamically adjusting the focus on different features, thereby significantly improving entity recognition accuracy. This provides new insights and methods for NER tasks in the ceramic field.