摘要
回答用户问询是电力营业厅的重要业务,使用智能系统识别用户意图可以大量减少人力成本、简化工作流程。真实的电力用户文本规模小、获取困难,从而导致智能系统深度学习效果仍待提升。为了解决语料不足的问题,提出一种基于对抗式生成网络的用户问询文本生成方法。实验通过将生成的文本加入训练集,使得RNN意图识别网络在测试集上的准确率由79.6%提升到82.1%。实验采用BLEU算法为评价方法,验证生成文本和真实文本的高相似度。由此证明使用用户问询文本生成模型可以生成符合实际需求的电力用户问询文本。
Answering user inquiries is an important business of the power business hall.Using intelligent systems to identify user intent can greatly reduce labor costs and workflow.The real power user text size is small and the access is difficult,which leads to the deep learning effect of the intelligent system still needs to be improved.In order to solve the problem of insufficient corpus,this paper proposes a user inquiry text generation method based on the confrontation generation network.The experiment adds the generated text to the training set,so that the accuracy of the RNN intent recognition network on the test set is increased from 79.6%to 82.1%.The experiment uses the BLEU algorithm as the evaluation method to verify the high similarity between the generated text and the real text.This proves that using the user inquiry text generation model can generate power user inquiry text that meets the actual needs.
作者
俞畅
欧阳昱
张波
刘辉舟
Yu Chang;Ouyang Yu;Zhang Bo;Liu Huizhou(School of Software Engineering,University of Science and Technology of China,Hefei 230031,China;State Grid Anhui Electric Power Company,Hefei 230061,China)
出处
《信息技术与网络安全》
2019年第11期67-72,共6页
Information Technology and Network Security
关键词
对抗式生成网络
文本生成
强化学习
generative adversarial network
text generation
reinforcement learning
