Knowledge Graph-based Enhancement Of Schizophrenia Symptom Recognition

As one of the important clinical information,the fine-grained classification identification of schizophrenia symptoms is important for similar medical record recommendation systems,disease research,and so on.However,most current symptom recognition tasks consider symptoms as a type of medical entity,and there is a lack of research on more fine-grained classification and recognition of symptom entities.Moreover,schizophrenia symptom entities are different from other types of diseases,most of them have prominent characteristic vocabulary but diverse expressions,long text spans,and confusing types,which make it difficult to classify and identify schizophrenia symptom entities.Based on this,this thesis explores a knowledge graphenhanced approach for the recognition of complex entities in schizophrenia from the feature vocabulary of schizophrenia symptom entities.This thesis constructs a knowledge map of schizophrenia symptoms and proposes a Biaffine model based on Knowledge-enabled Bidirectional Encoder Representation from Transformers(BK-BERT).The model incorporates information about schizophrenia symptom entities into the model in two ways: on the one hand,it is incorporated into the K-BERT(Knowledge-enabled Bidirectional Encoder Representation from Transformers)model in the form of a knowledge map to obtain a more easily understood.On the one hand,it is incorporated into the K-BERT model as a knowledge graph to obtain a collection of word vectors that are easier to understand.On the other hand,to solve the problems of a long span of entities and confusing types,a feature biaffine layer is added to the K-BERT model to improve the recognition capability of the model.Finally,the effectiveness of this model in the field of schizophrenia is demonstrated by comparing it with the baseline model.To better solve the problem of the long span of entity texts,this thesis finds that the feature vocabulary in its schizophrenia symptom knowledge map has more obvious location features in the entity when analyzing the entity data features.Therefore,based on BK-BERT,a symptom recognition model incorporating information on feature offset(BK-BERT-Offset)is proposed.The model further improves the interval delineation of entities by using the information on feature offset,which in turn improves the recognition ability of the model.Finally,the effectiveness of the model was verified through comparative experiments,especially in the case of difficult-to-identify symptom types such as delusions and hallucinations.