家禽诊疗文本多实体关系联合抽取模型研究

针对传统实体关系抽取方法中主体特征与句向量难以有效融合、现有BIO标注策略难以有效处理重叠关系的问题,提出一种基于BERT和双重指针标注的家禽疾病诊疗文本实体关系联合抽取模型(Joint extraction of entity relationship of poultry disease diagnosis and...

Joint Extraction Model of Multi-entity Relations for Poultry Diagnosis and Treatment Text

Aiming at the problems that the subject feature and sentence vector in the traditional entity relationship extraction method are difficult to effectively integrate, and the existing BIO annotation strategy is difficult to effectively deal with the overlapping relationships, a joint extraction of entity relationship of poultry disease diagnosis and treatment text (JEER_PD) based on BERT and dual-pointer was proposed. JEER_PD used the dual-pointer labeling (DPL) strategy to establish two pointer labelers at the head and tail, marking the beginning and ending positions of all entities at once; introduced the conditional layer normalization (CLN) network layer to strengthen the connection between the subject extraction task and the object relationship joint extraction task; and used the probability balance strategy (PBS) to combat the imbalance of positive and negative labels to accelerate the model convergence.The experimental results showed that the accuracy, recall and F1 value of JEER_PD were 97.69%, 97.59% and 97.64%, respectively, and the three indicators were significantly improved compared with that of the existing methods, which proved that JEER_PD can quickly and accurately extract the entity relationship triples in the complex knowledge text of the diagnosis and treatment of poultry diseases.