Development Of Intelligent Question-Answering System For Maize Breeding Based On Knowledge Graph

Maize breeding has made an important contribution to ensuring national food security.Among the many factors of maize yield increase,the contribution of variety improvement is about 35%,therefore,the selection and promotion of new varieties is a key factor to promote the continuous improvement of maize yield.In recent years,although the number of validated maize varieties in China has increased dramatically,generating a huge amount of information resources,these information are mostly presented in semi-structured and unstructured forms in web pages and books,and the information is scattered and the query efficiency is low by traditional search engine methods,which brings a lot of inconvenience to maize breeding work.In order to solve the above problems and help maize breeders to get valuable information quickly and accurately,this study used knowledge graph to organize data in the field of maize breeding and investigated intelligent question and answer methods based on deep learning,and based on this,designed and implemented an intelligent question and answer system for maize breeding.The main work and results accomplished in this paper are as follows.(1)Maize breeding knowledge graph construction and knowledge evolution.A joint BERT-CRF maize breeding entity relationship extraction method with embedded lexical information was designed for the problems of overlapping triads and diverse entity expressions in maize breeding data.The method adopted the strategy of simultaneous annotation of entity boundaries,relationship categories and entity location information to embed maize breeding vocabulary information in the BERT-CRF model,and designed the entity relationship triad matching algorithm ERTM to obtain the triads;To address the problem of data redundancy in the knowledge graph,a knowledge alignment method combining the edit distance algorithm and Jaccard similarity coefficient was designed.To address the incompleteness of the knowledge graph,a knowledge completion method Trans NWT based on entity neighborhood and weak translation was constructed.The experimental results showed that the F1 value of the knowledge extraction model constructed in this paper was 93.80%,and the accuracy of extracting the entity relationship triad of maize breeding was 95.88%,the Hit@10 of knowledge completion method was78%.(2)Research on intelligent question and answer method for maize breeding based on knowledge graph.To address the problem of inaccurate recognition of user question entities,a rule-based AC multi-pattern matching algorithm and a BERT-CRF model with embedded lexical information were used to extract question entities,and a string matching-based method was used to complete entity linking;To address the problem of varying question lengths,many intention categories and sparse features,a question intention understanding model combining BERT and RCNN was constructed;the question entities and intentions were parsed into Cypher query statements to complete answer queries.The experimental results showed that the F1 value of the question-intent understanding model constructed in this paper was 95.48% and the answer query accuracy was 86.31%.(3)Design and implementation of intelligent Q&A system for maize breeding based on knowledge graph.In response to the problem that maize breeders have difficulty in obtaining maize breeding information quickly and accurately,an intelligent question and answer system based on maize breeding knowledge graph was designed and implemented.The system taked the maize breeding knowledge graph as the knowledge base and the question and answer method based on the knowledge graph as the core,and realized the maize breeding entity query,relationship query and knowledge question and answer functions.The system can provide more accurate and effective information support for maize breeding work and improve the efficiency of breeders.