摘要
为提取文本的局部最优情感极性、捕捉文本情感极性转移的语义信息,提出一种基于卷积注意力机制的神经网络模型(CNN_attention_LSTM)。使用卷积操作提取文本注意力信号,将其加权融合到Word-Embedding文本分布式表示矩阵中,突出文本关注重点的情感词与转折词,使用长短记忆网络LSTM来捕捉文本前后情感语义关系,采用softmax线性函数实现情感分类。在4个数据集上进行的实验结果表明,在具有情感转折词的文本中,该模型能够更精准捕捉文本情感倾向,提高分类精度。
To extract the local optimal sentiment polarity of text and capture the semantic information of text sentiment polarity transfer,a neural network model(CNN_attention_LSTM)based on convolution attention mechanism was proposed.The convolution operation was used to extract the text attention signal and it was weighted into the Word-Embedding text distributed representation matrix,to highlight the emotional words and the turning words of the text focus.The long and short memory network LSTM was used to capture the emotional semantic relationship before and after the text.The softmax linear function was used to implement emotion classification.Results experiments on five data sets show that in texts with emotional turning words,the model can capture the emotional sentiment of the text more accurately,thus improving the classification accuracy.
作者
顾军华
彭伟桃
李娜娜
董永峰
GU Jun-hua;PENG Wei-tao;LI Na-na;DONG Yong-feng(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China;Hebei Provincial Key Laboratory of Big Data Computing,Hebei University of Technology,Tianjin 300401,China)
出处
《计算机工程与设计》
北大核心
2020年第1期95-99,共5页
Computer Engineering and Design
基金
国家青年科学基金项目(61806072)
天津市应用基础与前沿技术研究计划基金项目(16JCYBJC15600)
关键词
情感分析
注意力机制
自然语言处理
长短期记忆网络
深度学习
sentiment analysis
attention mechanism
natural language processing
long short-term memory
deep learning