摘要
为让用户通过输入自然语言就可以跟机器进行交互,实现文本的智能问答,提出基于混合神经网络的智能问答算法。将LSTM(long short-term memory)和CNN(convolutional neural network)相结合。利用LSTM计算问题和答案的语义特征,针对语义特征的选择进行改进。采用CNN对LSTM得到的语义特征进行筛选;通过计算问题和答案特征之间的相似度得到该模型的目标函数,给出问题对应的正确答案。仿真结果验证了该算法的可行性及有效性。
To make people interact with the machine possible by inputting the natural language and to realize the intelligent question-answering of the text,LSTM_CNN based on the hybrid neural network was proposed.The LSTM(long short-term memory)and CNN(long short-term memory)were combined.The semantic features of the questions and answers were calculated using the LSTM.The selection of features was then improved.The semantic features of the LSTM calculations were fina-lized using the CNN.The similarity between the problem feature and the answer one was evaluated,and the objective function of the model was obtained,thus the correct answer corresponding to the problem was given.The feasibility and effectiveness of the algorithm were verified by simulation experiments.
作者
付燕
辛茹
FU Yan;XIN Ru(College of Computer Science and Technology,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《计算机工程与设计》
北大核心
2020年第5期1434-1438,共5页
Computer Engineering and Design
基金
陕西省教育厅专项科学研究基金项目(16JK1505)。
关键词
智能问答
结巴分词
长短期记忆网络
卷积神经网络
深度学习
intelligent question-answering
Jieba
long short-term memory
convolutional neural network
deep learning
