Construction And Correlation Study Of Knowledge Graph Of Bovine Diarrheal Diseases

Bovine viral diarrhea virus(BVDV)is one of the main causative viruses of bovine diarrhea disease.BVDV can cause diarrhea,abortion,and immune suppression in cattle,resulting in significant economic losses in the global livestock industry.Preventing BVDV is one of the important measures to prevent bovine diarrhea disease,but the pathogenic and immune mechanisms of BVDV are still unclear.The construction of a knowledge graph that represents the interaction between BVDV and related genes can provide clues for understanding the interaction between BVDV and host cells,thus providing insights for the prevention and treatment of bovine diarrhea disease.Traditional methods for constructing knowledge bases require manual selection from various databases,which is inefficient and lacks scalability.Moreover,manually constructed knowledge bases typically use relational databases to store data,which is not suitable for multi-relational modeling and does not facilitate semantic information retrieval.Therefore,this thesis focuses on the intelligent construction methods and processes of knowledge graphs in the biomedical field,based on natural language processing technology,significantly improving the efficiency of knowledge graph construction in this domain.The main contributions of this thesis are as follows:1.Developed a data retrieval engine for the PubMed biomedical literature database.Addressing the issue of automated data retrieval,biomedical literature data is obtained through web crawling techniques by inputting keywords.2.Proposed a biomedical named entity recognition method that combines the BioBERT-BiLSTM-CRF model with a domain dictionary.Addressing the limitations of previous NER methods in biomedical text,such as poor recognition performance and cumbersome data annotation,this approach incorporates the BioBERT pre-trained language model specifically designed for the biomedical domain.It combines the BiLSTM-CRF model and leverages a domain dictionary to recognize biomedical entities.The proposed method achieved F1 scores of 89.02%,93.76%,86.47%,86.84%,and80.93% on the NCBI-Disease,BC5CDR-Chem,BC5CDR-Disease,BC2GM,and JNLPBA datasets,respectively.3.Proposed a biomedical relation extraction method that combines the PubMedBERT-BiGRU-Softmax model with rule templates.Addressing issues such as sample imbalance and noise interference in previous biomedical relation extraction tasks,this approach incorporates the PubMedBERT pre-trained language model designed for the biomedical domain.It combines the BiGRU-Softmax model and leverages rule templates to extract entity relationships.The proposed method achieved F1 scores of76.58% and 82.93% on the CHEMPROT and DDI datasets,respectively.4.Constructed a knowledge graph representing the interaction relationships between BVDV and genes.To overcome the limitations of relational databases in modeling multiple relationships and facilitating semantic information retrieval,a Neo4j graph database was used to store and construct the knowledge graph in the biomedical domain.5.Developed an interactive biomedical text mining system.Addressing the issue of inconvenient extraction of relevant textual information in the biomedical domain,an interactive text mining system was developed,enabling efficient organization of knowledge in the biomedical field.