Study On Entity Classification And Relation Extraction Of Rice Phenomics Knowledge Graph

With the continuous innovation and development of computer technology,the new generation of artificial intelligence technologies such as natural language processing,Internet of things,big data and other technologies have been widely used in the agricultural field,constantly promoting the rapid development of agricultural informatization.However.the data information in the field of rice phenomics is characterized by wide distribution,complex relationships and strong expertise.How to integrate and manage these data is extremely challenging.To this end,the paper will be introduced to the knowledge graph technology innovation in rice field of phenomics.through knowledge graph related technology to extract rice phenomics information and their relations,and visually describe their relationship in the form of a graph,so as to manage the rice phenomics data,and to dig deeper into the relationships between entities.In this paper,the knowledge graph of rice phenomics was constructed based on the knowledge of rice phenomics,the graph database Neo4j was used as the storage method of rice phenomics entities and relationships,at the same time,an efficient entity classification and relationship extraction method are proposed,It is beneficial to the construction of knowledge graph of rice phenomics and the application of agricultural information and intelligence.Specific research contents are as follows:(1)A combination model of rice phenomics entity classifier based on Stacking ensemble learning was proposed.Through the analysis of the traditional classification algorithm,it is found that the traditional classification algorithm can not satisfy the efficient classification of rice phenomics entities,especially the rice phenotypic data has the characteristics of large amount of data,complex and professional.In view of the low classification efficiency and limited improvement space of the traditional classification algorithm,this paper proposes a classifier combination model based on Stacking ensemble learning by using the two-layer Stacking framework based on Stacking ensemble learning strategy,combined with TF-IDF technology and LSI model to preprocess the rice phenomics data.The results showed that the combination model of rice phenomics entity classifier based on Stacking ensemble learning could significantly improve the classification efficiency,and the accuracy of entity classification was high,with an average accuracy of 6.78%higher than that of traditional classifier.(2)A deep learning relationship extraction model based on Attention mechanism is proposed.Traditional relationship extraction methods are feature-based methods and kernel-based methods,which usually rely on natural language processing tools,and these tools are prone to errors,which will lead to a low accuracy of the extraction task and seriously affect the performance of the relationship extraction system.Deep learning algorithm not only does not need natural language processing tools,but also no longer needs to use artificial form for feature establishment.In addition,the Attention mechanism can force the model to focus on important parts and increase the weight of effective samples.In this paper,the Attention-mechanism-based deep learning algorithm model and related algorithms are studied,and applied to the relation extraction of rice phenomics texts,so that it can efficiently complete the extraction of entity relations in the field of rice phenomics.(3)To construct the domain knowledge graph of rice phenomics.In this paper,the national rice data center website and Hudong Wiki website are used as the main data sources,and Scrapy crawler framework is used to extract rice phenomics information from the above data sources.The knowledge base of rice phenomics was constructed by analyzing the knowledge of rice phenomics,dividing and extracting the relations of entities in this field.Meanwhile,a combination model of entity classifiers in the field of rice phenomics based on ensemble learning was introduced to classify the entities in the field of rice phenomics,and a deep learning model was used to relation extraction.Finally,in order to solve the problem of knowledge storage in relational database,the paper used Neo4j a graph database to store the knowledge graph data of rice phenomics in the form of attribute map.Moreover,the development of knowledge mapping system of rice phenomics was completed,and the application of knowledge graph technology in agriculture was realized.(4)Functional realization of knowledge graph system in the field of rice phenomics.Through requirements analysis and architecture design,the Django framework was used to complete the establishment of the knowledge graph system in the field of rice phenomics,and realized the functions of entity identification,entity query,relational query,knowledge classification and knowledge visualization.An example shows that the knowledge graph system can supplement and repair the knowledge of rice phenomics,query the knowledge accurately and display it visually,which verifies the reliability and practicability of the system.