面向开放域深度学习的自动问答技术研究  

Research on automatic question and answer technology for open domain deep learning

作　　者：莫丽娅[1] MO Liya(Xi’an Fanyi University,Xi’an 710105,China)

机构地区：[1]西安翻译学院,西安710105

出　　处：《自动化与仪器仪表》2023年第5期216-220,共5页Automation & Instrumentation

基　　金：陕西省教育科学“十四五”规划2021年度课题《英语思辨阅读与课程思政融合路径研究》(SGH21Y0456);2021年校级《基础英语III、IV线上线下混合式课程》(ZK2122)。

摘　　要：结合聊天机器人背景,提出一个面向开放域深度学习的人机交互英语自动问答系统。首先,通过问答系统子模块建立问答库和收录问题预处理;然后利用机器学习算法进行关键词扩展、问题分类、相似度计算和答案匹配抽取;最后采用基于LSTM的Seq2seq模型实现英语聊天机器人,并在其基础上加入注意力机制和集数搜索算法,以提升系统自动问答质量。结果表明,相较于RNN和GRU神经元生成回复,LSTM神经元的生成回复结果更加准确。且添加注意力机制和集数搜索后,模型收敛速度显著提升。系统测试发现,英语问答系统子模块和英语聊天机器人的问答正确率分别为95.48%和96.52%,系统自动问答正确率为96%。由此可知,本系统可实现人机交互和英语问题的自动问答。Based on the background of chat robot,a human-computer interactive English automatic question answering system for open domain deep learning is proposed.First of all,the question-answering system sub-module is used to establish the question-answering database and pre-process the included questions;Then the machine learning algorithm is used for keyword expansion,question classification,similarity calculation and answer matching extraction;Finally,the Seq2seq model based on LSTM is used to implement the English chat robot,and the attention mechanism and set number search algorithm are added to improve the quality of the system's automatic question answering.The results showed that compared with RNN and GRU neurons,LSTM neurons produced more accurate results.After adding attention mechanism and set number search,the convergence speed of the model is significantly improved.The system test found that the correct rate of the English question answering system and the English chat robot were 95.48% and 96.52% respectively,and the correct rate of the automatic question answering system was 96%.Therefore,this system can realize human-computer interaction and automatic question answering of English questions.

关 键 词：开放域 人机交互 Seq2seq模型 自动问答系统 聊天机器人 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]