基于深度学习的多维特征微博情感分析

Analysis of Weibo sentiment with multi-dimensional features based on deep learning

提出1种基于卷积神经网络的多维特征微博情感分析新机制;利用词向量计算文本的语义特征,结合基于表情字符的情感特征,利用卷积神经网络挖掘特征集合与情感标签间的深层次关联,训练情感分类器;结合微博文本的语义和情感特征,同时利用卷积神经网络的抽象特征提取能力,进而改善情感分析性能。研究结果表明:引入表情字符的情感特征模型可使情感分析准确率提高2.62%;相比基于词典的机器学习模型,新机制将情感分析准确率与F度量分别提升21.29%和19.20%。

A new mechanism of Weibo sentiment analysis based on convolutional neural networks with multidimensional features was proposed. The proposed mechanism combines semantic features from word vectors with sentiment features from emoticons, in which convolutional neural networks was used to mine deep correlation between features and labels. The performance of Weibo sentiment analysis was improved through mining multi-dimensional features and utilizing abstract features extraction ability of convolutional neural networks. The results show that the accuracy of sentiment analysis model based on emoticons increases by 2.62%. The accuracy and F measure increase by 21.29% and 19.20% respectively compared with that of machine learning model based on lexicon.