基于多注意力CNN的问题相似度计算模型  

Question Similarity Calculation Model Based on Multi-Attention CNN

作　　者：冯兴杰[1] 张乐 曾云泽 FENG Xingjie;ZHANG Le;ZENG Yunze(College of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院

出　　处：《计算机工程》2019年第9期284-290,共7页Computer Engineering

基　　金：国家自然科学基金青年科学基金(61301245,61201414);赛尔网络下一代互联网技术创新项目(NGII20160605)

摘　　要：在智能客服问答系统中,用户所提问句具有咨询意图复杂、上下文相关性弱以及口语化等特点,导致问句相似度计算的准确率不高,出现答非所问的情况。提出一种基于卷积神经网络的相似度计算模型MA-CNN。通过2个不同的注意力机制,同时关注词汇间的语义信息和句子间的整体语义信息,提高智能客服对问题的理解能力。实验结果表明,与基于词向量和基于循环神经网络的模型相比,MA-CNN模型对问句的辨识能力更强,其 F 1值最高可达0.501。In the intelligent customer service question-answering system,the questions asked by users are characterized by complex consultation intention,weak contextual relevance,and serious colloquialization.As a result,the accuracy of question similarity calculation is not high and irrelevant answer occurs.In order to solve these problems,a similarity calculation model MA-CNN based on Convolutional Neural Network (CNN) is proposed.Focusing on the semantic information between words and the overall semantic information between sentences through two different attention mechanisms,the problem understanding ability of the intelligent customer service can be improved.Experimental results show that,compared with the models based on word vector and Recurrent Neural Network(RNN),the MA-CNN model has stronger ability to identify questions,and its F 1 value can reach up to 0.501.

关 键 词：智能客服 文本相似度 词语语义 句子语义 卷积神经网络 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]