基于深度学习的机械智能制造知识问答系统设计  

Design of knowledge question-answering system for mechanical intelligent manufacturing based on deep learning

作　　者：朱建楠 梁玉琦 顾复[4] 郭剑锋[5] 顾新建 ZHU Jiannan;LIANG Yuqi;GU Fu;GUO Jianfeng;GU Xinjian(Institute of Automatic Control,Lanzhou Jiaotong University,Lanzhou 730070,China;Key Laboratory of Plateau Traffic Information Engineering and Control of Gansu Province,Lanzhou 730070,China;Key Laboratory of Opto-Technology and Intelligent Control,Ministry of Education,Lanzhou 730070,China;School of Mechanical Engineering,Zhejiang University,Hangzhou 310027,China;Institute of Science and Development,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]兰州交通大学自动控制研究所,甘肃兰州730070 [2]甘肃省高原交通信息工程及控制重点实验室,甘肃兰州730070 [3]光电技术与智能控制教育部重点实验室,甘肃兰州730070 [4]浙江大学机械工程学院,浙江杭州310027 [5]中国科学院科技战略咨询研究院,北京100190

出　　处：《计算机集成制造系统》2019年第5期1161-1168,共8页Computer Integrated Manufacturing Systems

基　　金：国家重大专项子课题资助项目(2016ZX05040-001);国家自然科学基金面上资助项目(7167010907,71271200,51775493)~~

摘　　要：为了构建智能制造知识问答系统,促进智能制造知识传递,加快智能制造产业布局,利用深度学习算法对传统问答系统构建流程过于复杂、所需手工与先验知识要求过高、问题与答案无法有效映射等问题进行改进。采用长短记忆神经网络算法来避免一般深度学习算法在进行梯度优化时的梯度消失与梯度爆炸问题,算法中的门机制能够消除链式法则对梯度过度优化的影响,直接对句子的语义做出解析,并利用相似度计算判别回答的正确与否。通过在评测集上的验证实验表明,该语义解析方法能够显著提升问答系统的准确率。To construct the intelligent manufacturing Question-Answering (QA) system, facilitate the knowledge sharing of intelligent manufacturing and accelerate the intelligent manufacturing industrial layout, the deep learning algorithm was used to improve the problems that the construction process of the traditional QA System was too complicated, the requirement of human intervention and prior knowledge was too high, the questions and answers could not be effectively mapped and so on. Long Short-Term Memory (LSTM) deep neural network algorithm was applied to avoid the gradient disappearance and gradient explosion problems of the general deep learning algorithm. The gate mechanism in LSTM algorithm could avoid the influence of chain rule on gradient over-optimization and analyze the semantics of sentence directly. The similarity calculation was used to determine the correct answer or not. The verification experiments on the validation set showed that the proposed method of semantic analysis had a clear improvement in the accuracy of QA System.

关 键 词：深度学习 智能制造 Encoder-Decoder框架 问答系统 长短记忆神经网络 门机制 相似度 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]