基于卷积神经网络的在线评论情感分析模型  

Sentiment Analysis of Online Comments Based on Convolutional Neural Network

作　　者：毕殿杰[1] 魏苏林[1] 赵涛[1] 张子振[1] BI Dianjie;WEI Sulin;ZHAO Tao;ZHANG Zizhen(School of Management Science and Engineering,Anhui University of Finance and Economics,Bengbu Anhui,233030,China)

机构地区：[1]安徽财经大学管理科学与工程学院

出　　处：《河北科技师范学院学报》2019年第2期41-47,共7页Journal of Hebei Normal University of Science & Technology

基　　金：安徽财经大学校级科学研究项目(项目编号:ACKY1857);安徽省自然科学研究基金项目(项目编号:1608085QF145)

摘　　要：针对标准卷积神经网络在文本情感分析过程中忽略了句子的整体结构信息的缺陷,本次研究在卷积神经网络的输入端加入注意力机制,提出了基于双通道输入的分段池化卷积神经网络模型(AF_CNN模型),该模型既能够有效提取文本局部最优特征,又能够捕捉到上下文词语之间的相关性。针对体育新闻评论情感分析的实验结果表明,与标准的卷积神经网络模型相比,本次研究提出的AF_CNN模型在分类准确率、召回率和F1值等评价指标上,分别提升了3.40%,0.47%,1.96%。In view of the fact that the standard convolution neural network ignores the overall structure information of sentences, this paper adds attention mechanism to the input of convolution neural network, and proposes a two-channel input-based Chunk-Max pooling convolution neural network model (AF_CNN model), which can not only effectively extract the local optimal features of text, but also capture the correlation between context words. The experimental results of sentiment analysis of sports news commentary show that compared with the standard convolutional neural network model, the AF_CNN model proposed in this paper improves the classification accuracy, recall rate and F1 value by 3.40%, 0.47% and 1.96% respectively.

关 键 词：在线评论 情感分析 卷积神经网络 注意力机制 分段池化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]