Research And Application Of Named Entity Recognition And Knowledge Graph Construction For Pig Disease

At present,pig breeding has become one of the important industries in the world.However,due to the expansion of breeding scale and intensive breeding,various diseases have become important constraints on pig production.For breeding personnel,mastering the prevention and treatment technology of pig disease and taking effective prevention and control measures are important guarantees for the healthy development of pig industry.Because of the wide range of knowledge sources in the field of swine diseases and the various ways of data storage and presentation,it is difficult for breeders to obtain the relevant disease knowledge efficiently and quickly.With the rapid development of artificial intelligence and big data technology,natural language processing technology and knowledge map construction technology are widely used in various fields.Knowledge extraction and knowledge map construction oriented to pig disease field are expected to solve this problem.In this thesis,the lack of knowledge in the field of porcine disease is taken as the starting point,and the focus is on domain named entity recognition.By using other key technologies in the construction of knowledge map,the thesis studies how to build high-quality knowledge map in the field of porcine disease,so as to provide data support for porcine disease knowledge question answering system,so as to realize accurate question answering of porcine disease knowledge.The main work of this thesis is as follows:(1)To study and construct a pig disease corpus.In response to the lack of a publicly available corpus in the field of pig disease,this article constructs a named entity recognition corpus in the field of pig disease.More than 370000 words of text in the field of pig disease were obtained from multiple data sources.After preprocessing operations such as corpus cleaning and labeling,a total of 10783 entities of 11 types and 1696 related disease images were obtained.The construction of this dataset and the acquisition of images provide data support for named entity recognition research in the field of pig disease and the construction of a multimodal knowledge graph of pig disease.(2)The named entity recognition algorithm in pig disease field was studied.There are many professional terms,complex sentence structure,long names of related entities,and the mixed use of numbers,letters and special characters in the field of swine disease.This article proposes a named entity recognition model based on CNN and multi head self attention mechanism,which uses pre trained models to obtain word vector representations and then inputs CNN to obtain more local context features.Then,BILSTM is used to process the local features extracted from the CNN layer to make it more global.The multi head self attention mechanism is used to obtain the correlation strength and global feature representation between the CNN output information.Finally,combine the outputs of the above two modules and send them to the CRF layer to obtain the most likely global optimal label sequence.The results show that the method proposed in this article performs better than other methods.(3)To study and construct multimodal knowledge graph in pig disease field.This article uses the named entity recognition algorithm mentioned above to obtain different types of entities for the 11 defined entity types,and classifies the extracted entities by predefined 10 relationship types.Knowledge fusion is performed on multi-source data,and the Neo4 j graph database is used to store pig disease knowledge.A multi-modal knowledge graph in the field of pig disease is constructed,laying the foundation for downstream knowledge graph question and answer systems.(4)Design and implement pig disease knowledge service system.Based on the above research work,the implementation of pig disease knowledge question and answer mainly includes question entity recognition,question classification,graph query,etc.Users input questions through the system,and the system uses deep learning models to identify entities in question sentences,and classifies questions based on question keywords.Then,the system maps the classification results into a Cypher query template and generates query statements,queries the answers in the knowledge graph and returns the output,ultimately achieving graph visualization and knowledge question and answer.