Construction of Knowledge Graph of HIV-associated Neurocognitive Disorders Syndrome based on Deep Learning

If HIV-associated Neurocognitive Disorder (HAND) can be diagnosed and treated early, it may delay or reverse its pathological process and improve the survival rate of patients. At present, there is little statistical information about HAND, which is very disadvantageous to the prevention and treatment of HAND. Therefore, this paper synthetically uses deep learning models such as bidirectional LSTMs, conditional random fields and PCNN to carry out entity recognition and relationship extraction for text data, such as electronic medical record and medical community, to construct visual knowledge graph. Firstly, entity type and relation type are defined, and then multi-source data are fused, and then entity recognition of BIO annotated data sets is carried out by using the BERT-BiLSTM-CRF model. It is found that the effect of using the BERT pretraining model is better than word2vec; then, the neural network PCNN-Attention based on sentence level selective attention mechanism is used. It is found that the precision rate, recall rate and F1 value of the model are better than PCNN-ONE and PCNN-AVE models. Finally, the entity and entity relationship are visualized by using Neo4j graph database. In this experiment, the HAND related knowledge graph was constructed and visualized, which is helpful to the popularization of HAND related medical knowledge and the diagnosis of doctors, and it is helpful to the early detection of ANI, and plays an important role in delaying the pathology.