Research On Knowledge Graph Construction Technology For Husbandry

Husbandry is the pillar industry of China’s rural economy and the nexus industry of agricultural restructuring.With the development of the information age,data generation in the husbandry industry has accelerated significantly.Making full use of the knowledge of husbandry contained in the massive amount of data can provide help to the information intelligence of China’s husbandry industry.There are problems such as strong professionalism and difficulties in sharing knowledge in the field of husbandry.Therefore,the high-quality extraction and integration of husbandry knowledge has become the focus of research.The knowledge graph can use a unified structure to express unstructured information in the field of husbandry,form a network knowledge structure,and reduce the application threshold of husbandry knowledge.The construction of husbandry knowledge graph can provide knowledge base support for subsequent application scenarios such as intelligent consultation,intelligent question and answer,and decision-making system.According to different fields and uses,knowledge graphs can be divided into general knowledge graphs and domain knowledge graphs.Among them,the domain knowledge graph pays more attention to depth.Due to its strict requirements for professional knowledge reserves,there are problems of over-reliance on experts and high construction costs.Therefore,the automatic construction of knowledge graphs is the current mainstream method of domain knowledge graph construction.Knowledge extraction in automatic construction technology is the key to constructing knowledge graph of husbandry.This research,as a sub-task of husbandry knowledge graph construction,is responsible for building an overall framework,constructing husbandry knowledge graph through four steps of data collection,knowledge extraction,knowledge modeling,and knowledge storage.The knowledge involved includes husbandry species,veterinary diseases,and veterinary drugs.The husbandry knowledge graph was constructed,and the method research was carried out for the named entity recognition task in the construction process.The main research results are as follows:(1)Collections of multi-source data.This method was based on the 33 types of livestock breeds in the National Inventory of Livestock and Poultry Genetic Resources,semi-structured and unstructured data were obtained from data sources such as the Livestock and Poultry Genetic Resources Census Information System,the National Veterinary Drug Basic Database,and veterinary monographs,the collection content involved husbandry species,veterinary diseases,and veterinary drugs,and veterinary diseases,formed a husbandry data set.(2)Research on named entity recognition methods,and a Chinese named entity recognition method based on matching word weight optimization was proposed.The method made full use of the dictionary matching feature.First,the vector representation and part-of-speech tagging of each character were obtained by using the pre-trained model and word segmentation tools;then the potential phrases were matched in the dictionary,and the phrases were weighted according to the optimized weights of the matching word frequency and document count.Combined with the character vector,the multi-feature representation of the character was obtained;finally,the bidirectional long short-term memory network was used for training,and the conditional random field was used to complete the label inference to obtain the entity.The experimental results shown that the method is better than the comparison model on the public dataset and the self-built Chinese animal disease entity dataset.(3)A husbandry knowledge graph has been built using a top-down approach.The seven-step ontology construction method and the Protégé ontology construction tool were used to define the conceptual architecture of husbandry knowledge and construct the husbandry domain ontology.Extracted 8827 entities such as veterinary diseases and clinical symptoms using the proposed entity identification method,3102 entities such as livestock breeds and animal classification,17760 attribute triples and 10110 relationship triples were extracted using the rule-based method.A Neo4 j graph database was used to complete the storage and visualization of the husbandry knowledge graph,with 6138 entities and 27870 triples stored.