Research On Automatic Text Extraction And Application Of Knowledge Graph For Diagnosis And Treatment Of Poultry Common Diseases

Poultry farming is the basic industry of rural economic development,China’s total poultry breeding and egg production ranked first in the world for many years.However,the domestic information technology and modern farming technology are still in the initial stage,and the quality of poultry products still has a certain gap compared with developed countries.Disease control is a key factor affecting the production and quality of poultry products in production practice.The variety of poultry diseases in China,the speed of updating,the wide range of morbidity,the difficulty of diagnosis and treatment,most farmers have difficulty in obtaining timely and high-quality knowledge of poultry disease diagnosis and treatment.To solve this problem,we combine knowledge mapping and natural language processing technology to automatically extract semistructured and unstructured poultry disease diagnosis and treatment knowledge from books and the Internet.After refining and integrating,we construct a knowledge map of common poultry disease diagnosis and treatment,and help farmers quickly obtain high-quality poultry disease diagnosis and treatment knowledge in the form of a knowledge quiz application.Firstly,we studied the joint extraction method of entity relations for poultry disease diagnosis and treatment texts,and realized the transformation from knowledge text to entity relations triad.Then,by modeling the conceptual and entity layers of the knowledge graph,we initially built a knowledge graph framework for the diagnosis and treatment of common poultry diseases.Finally,the question and answer algorithm that can accurately answer 13 types of diagnosis and treatment knowledge is constructed and implemented based on semantic parsing,and the common poultry disease diagnosis and treatment knowledge question and answer system is designed and implemented by using Flask lightweight web framework.The work content is as follows.(1)We propose a Joint Extraction model of Entity Relationship of Poultry Disease diagnosis and treatment text(JEER＿PD)based on the BERT language model.JEER＿PD uses Dual-Pointer Labeling(DPL)strategy to build two pointer labelers,head and tail,to label the start and end positions of all entities at once.A conditional layer normalization(CLN)network layer is added to the model to strengthen the connection between the subject extraction task and the object-relational joint extraction task;the probabilistic balancing strategy PBS is introduced to combat the positive and negative class label category imbalance and accelerate the model convergence.97.69%,97.59% and 97.64% of JEER＿PD accuracy,recall and F1 values are achieved respectively,which are significant improvements over existing methods.The experiments demonstrate that the model can quickly and accurately extract the triad of entity relationships in the complex knowledge text of poultry disease diagnosis and treatment.(2)We took the electronic books related to the diagnosis and treatment of common poultry diseases and the website data of "Veterinary Medicine Information Network" as the original data of the knowledge map,and used the JEER＿PD entity relationship extraction model to obtain the triad of entity relationships of diagnosis and treatment knowledge in batch.By analyzing the structure of poultry disease diagnosis and treatment knowledge and dividing the attribute labels for the entities and relations of the triad by category,the conceptual and entity-level modeling of the knowledge map was completed,and the knowledge map of common poultry disease diagnosis and treatment was initially constructed.In order to avoid the data redundancy problem caused by the repeated entry of entity relations in the knowledge map,the knowledge map is stored using the graph database Neo4 j and maintained by Cython coding.(3)In order to meet the demand of farmers for poultry disease diagnosis and treatment knowledge,we designed and developed a common poultry disease diagnosis and treatment knowledge quiz system based on the constructed knowledge map.The system has three functional modules: text information extraction,poultry disease knowledge quiz,and knowledge mapping tools,and has the advantages of simple operation,rapid response,concise interface,and easy learning and use.The system can visualize the treatment knowledge in the form of diagrams and tables from multiple angles to meet the knowledge needs of farmers in different application scenarios.Finally,the practicality and reliability of the system is verified by citing some usage examples.