基于卷积神经网络的情感分析算法  

EMOTION ANALYSIS ALGORITHM BASED ON CONVOLUTION NEURAL NETWORK

作　　者：李佳丽 封化民[1,2] 潘扬 徐治理 刘飚[2] Li Jiali;Feng Huamin;Pan Yang;Xu Zhili;Liu Biao(Xidian University,Xi'an 710071,Shaanxi,China;Beijing Electronic Science and Technology Institution,Beijing 100070,China)

机构地区：[1]西安电子科技大学,陕西西安710071 [2]北京电子科技学院,北京100070

出　　处：《计算机应用与软件》2018年第4期287-292,共6页Computer Applications and Software

基　　金：中央高校基本科研业务费专项资金资助课题(2017CL02)

摘　　要：随着词向量的广泛应用,情感词典在情感分析任务中不再使用。但是使用传统机器学习方法时,情感词仍然提供了重要的特征。通过结合词向量与情感词典,提出两种基于卷积神经网络的情感分析方法,分别为朴素连接法和独立卷积法。其中情感词典的构建采用传统的种子词方法,根据How Net和Word2Vec共同计算出当前语料库的词语-情感分数键值对。实验结果表明,提出的情感分析方法可以有效地提高情感分析的准确率。With the wide use of word vectors,the emotion dictionary is no longer used in emotion analysis tasks.However,emotional words still provide important features in traditional machine learning methods.By combining the word vectors with the emotion dictionary,this paper presents two emotion analysis methods based on convolution neural network,which are the simple connection method and the independent convolution method respectively.The construction of the emotion dictionary used the traditional seed word method,and according to HowNet and Word2Vec,the word-sentiment score key-value pair of the current corpus was calculated.Experimental results show that the proposed emotion analysis method effectively improves the accuracy of emotional analysis.

关 键 词：词向量 情感词典 卷积神经网络 朴素连接 独立卷积 HOWNET Word2Vec 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]