English

智能化农业装备学报(中英文) ›› 2024, Vol. 5 ›› Issue (4): 39-50.DOI: 10.12398/j.issn.2096-7217.2024.04.003

• • 上一篇    下一篇

基于知识图谱的农作物病虫害问答系统研究

赵泽行1(), 吴晓鹏1, 王怡馨2, 闫小丽1, 黄玉祥1, 高筱钧3,4()   

  1. 1.西北农林科技大学机械与电子工程学院,陕西杨凌,712100
    2.西北农林科技大学信息工程学院,陕西杨凌,712100
    3.南京农业大学工学院,江苏 南京,210031
    4.农业装备技术全国重点实验室,北京市,100083
  • 收稿日期:2023-12-27 修回日期:2024-03-21 出版日期:2024-11-15 发布日期:2024-11-15
  • 通讯作者: 高筱钧,男,1990年生,黑龙江齐齐哈尔人,博士,副教授;研究方向为高速精播技术与装备等。E-mail: gxj-1234@126.com
  • 作者简介:赵泽行,男,2001年生,河南许昌人;研究方向为智慧农业与深度学习。E-mail: 2019015213zzx@nwafu.edu.cn
  • 基金资助:
    农业装备技术全国重点实验室开放课题(NKL-2023-008);陕西省重点研发计划项目(2023-YBNY-204);中国博士后科学基金项目(2022M722609);陕西省博士后科研项目(2023BSHEDZZ149)

Research on question answering system for crop diseases and pests based on knowledge graph

ZHAO Zexing1(), WU Xiaopeng1, WANG Yixin2, YAN Xiaoli1, HUANG Yuxiang1, GAO Xiaojun3,4()   

  1. 1.College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling 712100,China
    2.College of Information Engineering,Northwest A&F University,Yangling 712100,China
    3.College of Engineering,Nanjing Agricultural University,Nanjing 210031,China
    4.National Key Laboratory of Agricultural Equipment Technology,Beijing 100083,China
  • Received:2023-12-27 Revised:2024-03-21 Online:2024-11-15 Published:2024-11-15
  • Contact: GAO Xiaojun

摘要:

在大数据时代的推动下,智能问答系统在各个领域得到广泛应用,为用户提供高效的答案,相比传统的文本知识收集和网络搜索引擎检索,具有明显优势。随着知识图谱技术的快速发展,智能问答系统迎来了新的发展阶段。本研究响应了智慧农业的需求,构建了一个基于知识图谱的农作物病虫害智能问答系统,旨在为用户提供与农作物病虫害相关的问答服务。主要工作包括:(1) 农作物病虫害数据获取:利用分布式爬虫框架爬取农作物病虫害相关网页数据,并进行数据清理、分析、结构化等预处理操作。(2) 知识图谱构建:对数据进行分析后,定义知识本体的实体类别与关系类别,完成知识图谱的模式层构建。利用基于规则的三元组模板对半结构化文本进行实体提取,构建数据层,并将三元组存储至Neo4j图数据库。(3) 问答算法设计:利用BERT-BiLSTM-CRF模型进行问句实体识别,BERT-RNN模型进行问题分类。匹配模板后通过Cypher语句进行查询,将答案处理为自然语言形式并返回。(4) 问答系统实现与可视化:结合农作物病虫害知识图谱与问答算法,使用Flask框架和多种web技术实现用户提问、实体识别、知识检索和答案返回等功能。试验结果表明,实体识别与问题分类模型的准确率(Precision)、召回率(Recall)和F1分别达到了93.22%、92.69%、92.21%和94.37%、93.92%、92.66%。与其他搜索途径相比,问答系统展现出了较高的准确性和稳定性。这项研究为农业信息化提供了一种智能化的解决方案,为用户获取农作物病虫害知识提供了新途径。

关键词: 农业信息化, 农作物病虫害, 知识图谱, 问答系统

Abstract:

Propelled by the era of big data, intelligent question-answering systems have found widespread applications across various domains, offering users a means to swiftly and efficiently acquire answers. Compared to traditional methods of collecting textual knowledge and utilizing web search engines, these systems demonstrate distinct advantages. With the rapid evolution of knowledge graph technology, intelligent question-answering systems have entered a new stage of development. In response to the call for smart agriculture, this study has undertaken the construction of a knowledge-based question-answering system based on knowledge graphs, aiming to provide users with question-answering services related to crop diseases and pests. The primary tasks include: (1) Acquisition of crop disease and pest data: Employing a distributed web crawling framework to retrieve data from web pages related to crop diseases and pests and preprocessing operations involve data cleaning, analysis, and structuring. (2) Construction of knowledge graph: Following data analysis, defining entity and relationship categories for the knowledge ontology and completing the pattern layer construction of the knowledge graph. Utilizing rule-based triple templates for extracting entities from semi-structured text, building the data layer, and storing triples in the Neo4j graph database. (3) Design of question-answering algorithms: Employing the BERT-BiLSTM-CRF model for question entity recognition and the BERT-RNN model for question classification. Matching templates, executing queries through Cypher statements, and processing answers into natural language forms for return. (4) Implementation and visualization of the question-answering system: Integrating crop disease and pest knowledge graphs with question-answering algorithms, the Flask framework and various web technologies are used to implement functionalities such as user questioning, entity recognition, knowledge retrieval, and answer presentation. Experimental results indicate that the entity recognition and question classification models achieved Precision (P), Recall (R), and F1 scores of 93.22%, 92.69%, 92.21%, and 94.37%, 93.92%, 92.66%, respectively. Compared to other search methods, the question-answering system demonstrated higher accuracy and stability. This study provides an intelligent solution for agricultural informationization, offering a new pathway for users to acquire knowledge on crop diseases and pests.

Key words: agricultural informationization, crop disease and pest, knowledge graph, question and answering system

中图分类号: