基于BERT_DPCNN的牛类疾病问句分类研究  

Classification of bovine disease questions based on BERT_DPCNN

作　　者：黄双斌 王梅嘉 高浏洋 HUANG Shuangbin;WANG Meijia;GAO Liuyang(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,西安710021

出　　处：《智能计算机与应用》2025年第3期140-144,共5页Intelligent Computer and Applications

基　　金：陕西省自然科学基金研究计划(2022JQ-175);陕西省教育厅科学研究计划项目(22JK0303)。

摘　　要：为帮助养殖户实现牛类养殖的精准、科学化管理,基于BERT、TextCNN、TextRNN模型,研究牛类疾病的问句分类方法,为构建面向牛类疾病的问答系统提供技术支撑。通过设计爬虫获取惠农网、百度贴吧等原始数据,并对数据进行预处理,获取了包含5056条数据的数据集,将数据进一步划分为定义、预防、病因、症状、治疗和诊断共6类,以构建牛类疾病分类语料库。实验表明,BERT模型在6类精度有4类不弱于其他模型,在不同大小的数据集上预训练BERT模型在加权F1值上均优于TextCNN和TextRNN模型,与BERT其他变种模型进行了实验对比,BERT_DPCNN模型比BERT模型加权F1值高0.3%,考虑问答系统问句分类精确度要求高,选取BERT_DPCNN模型作为问句分类模型。In order to help farmers achieve accurate and scientific management of cattle breeding,this paper studies the classification method of questions for cattle diseases based on BERT,TextCNN and TextRNN models,and provides technical support for the construction of a question-answering system for cattle diseases.The crawler is designed to obtain the original data of Huinong network and Baidu Post Bar,and the data is preprocessed to obtain a dataset containing 5056 pieces of data,and the data is further divided into 6 categories:definition,prevention,etiology,symptoms,treatment and diagnosis,so as to build a classification corpus of bovine diseases.Experiments show that BERT model is no weaker than other models in four of the six categories of accuracy.On data sets of different sizes,pre-trained BERT model is superior to TextCNN and TextRNN model in weighted F1 value.Experimental comparison is conducted with other BERT variant models.The weighted F1 value of BERT_DPCNN model is 0.3%higher than that of BERT model.Considering the high accuracy requirement of question classification in question answering system,BERT_DPCNN model is selected as the question classification model.

关 键 词：自然语言处理 BERT 牛类疾病 问答系统 问句分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]