Design And Research On Q&A System Based On Agricultural Knowledge Graph

With the advancement of agricultural modernization,knowledge service system based on agricultural field has become a hot issue in agricultural informatization research.The traditional way of information acquisition is mainly using search engine,which returns a large number of Web links.The answers are highly dispersed and show the characteristics of multi-source heterogeneity.It can not provide knowledge services for agricultural workers quickly and accurately.The development of domain knowledge graph provides a high-quality knowledge base for KGQA in specific fields.Based on the constructed agricultural knowledge graph,this paper studies the agricultural knowledge Q&A system.The main research contents are as follows:(1)A knowledge graph in the field of agriculture is built.Firstly,this paper uses the Scrapy framework to crawl the Web Agricultural text data,and realizes the large-scale collection of agricultural data.Then,the storage scheme of knowledge graph is designed according to the characteristics of processed data.Finally,Cypher is used to store the structured entity data into the Neo4 j,and a knowledge graph in the agricultural field with high data quality,wide coverage and clear hierarchy is constructed.(2)The Q&A algorithm based on knowledge graph is studied.Based on the constructed knowledge graph,the Q&A algorithm in this paper is divided into two sub tasks: named entity recognition and question classification.For the questions of entity to entity task,PBERT-BiLSTM-CRF named entity recognition algorithm is constructed to recognize the entities such as crops,diseases,pests and pesticides.On the basis of this model,the combined multi features of the word itself and the radical side are added to carry out the experiment.Finally,the experimental results show that the PBERT-BiLSTM-CRF model combined with multi features in this paper has better effect in the task of named entity recognition in the agricultural field than other common models,and has the characteristics of fast training speed and high accuracy.Aiming at the questions of symptoms input by users,This paper uses a self attention classification model based on BiLSTM.This model brings a simple way for embedding,which can see which specific parts of the sentence are encoded into the embedding.Through the classification experiment with the same data processed by other neural network models,the results show that the question classification model in this paper can accurately classify users’ questions about symptom.Questions classification expands the scope of the question answering system and improves the value of the agricultural Q&A system based on knowledge graph.(3)An agriculture Q&A system based on knowledge graph has been designed and realized.Through the analysis of system requirements,the design and test of system,when users input natural language questions related to pests and pesticides in the agricultural field,the system can accurately give the corresponding text answers and the visualization of the corresponding entity relationship like the knowledge graph.