| 摘 要: 中医实体识别是智能医学领域一项重要的基础任务,针对通用的实体识别模型忽略了中医实体之间强关联性的问题,文章提出了一种特征融合方法。以预训练BERT(Bidirectional Encoder Representation from Transformers)模型为基础,对输入信息进行特征提取后,进一步对每个实体特征与语句特征进行融合,以获得更丰富的上下文特征,进而增强模型对中医实体的提取能力。使用临床数据开展实验证明,该方法与其他模型相比,在草药和症状实体识别任务上获得了更高的F1分数,分别为94.90%和83.92%,能更准确有效地提取医案中的实体。 |
| 关键词: 特征融合;命名实体识别;BERT;中医 |
|
中图分类号: TP391
文献标识码: A
|
|
| Research on Entity Recognition of Traditional Chinese Medicine Case by Fusing Entity and Statement Feature Information |
|
WANG Feng1, CHEN Genlang2, WU Chuang1
|
(1.School of Artif icial Intelligence, Zhejiang Sci-Tech University, Hangzhou 310018, China;
2.School of Computer and Data Engineering, NingboTech University, Ningbo 315100, China)
202130504162@mails.zstu.edu.cn; cgl@zju.edu.cn; 202130504177@mails.zstu.edu.cn
|
| Abstract: Traditional Chinese Medicine(TCM) entity recognition is an important fundamental task in the field of intelligent medicine. Aiming at the problem that the general entity recognition model ignores the strong correlations between TCM entities, this paper proposes a feature fusion method. Based on the pre-trained BERT (Bidirectional Encoder Representations from Transformers) model, the method extracts features from the input information and further fuses each entity feature with statement features to obtain richer contextual features, thereby enhancing the model's ability to extract TCM entities. Experiments conducted on clinical data demonstrate that the method obtained higher F1 scores of 94.90% and 83.92% in the herb and symptom entity recognition tasks, compared to other models, allowing for more accurate and effective extraction of entities in medical cases. |
| Keywords: feature fusion; named entity recognition; BERT; traditional Chinese medicine |
