Named Entity Recognition in the Elderly Dietary Domain Based on the BERT-BiLSTM-CRF Model
Yong Ge, Shenghui Zhao, Shu Guo Zhao, Cuijuan Shang · 2024
Chronic diseases in the elderly are closely related to dietary habits, and scientifically informed nutrition has become a crucial component in strategies aimed at preventing chronic diseases and delaying aging. As a key technology, knowledge graphs can structure and process vast and diverse dietary data of the elderly, extracting nutrition knowledge closely linked to their health. This supports the formulation of personalized dietary plans and optimizes dietary structures and health management for the elderly. However, most existing elderly dietary data are unstructured, making the identification of named entities from large volumes of unstructured text data essential for the creation of knowledge graphs. In this study, we propose and construct a named entity recognition (NER) dataset specifically designed for the field of elderly dietary nutrition, named ElderDiet_NER. Based on this dataset, we introduce a model that combines Transformers (BERT), Bidirectional Long Short-Term Memory networks (BiLSTM), and Conditional Random Fields (CRF) for named entity recognition. When tested on the ElderDiet_NER dataset, the model achieved a precision of 91.51%, a recall of 91.47%, and an Fl-score of 91.49%. The experimental results demonstrate that this approach can effectively extract specific entity categories from text data related to elderly nutrition and diet, laying a solid foundation for the subsequent construction of knowledge graphs in the elderly dietary domain.