基于卷积神经网络的短评语情感分类

Classification of Short Comment Emotion Based on Convolutional Neural Network

随着社交网络以及电子商务的飞速发展,越来越多的用户习惯于在互联网上针对商品发表评论,造成各大电子商务网站上产品的短评语总量飞速上涨。面对海量内容相似、格式随意的评语,研究人员以及数据使用者仅凭人力在众多短评语中提取对自己有价值的信息比较困难,因此短文本评语的情感分类得到了广泛的关注。针对人工提取困难的问题,提出一种改进的卷积神经网络模型。该模型通过词嵌入和多通道卷积神经网络结合的方式实现了短文本评论的情感分类,弥补了支持向量机模型带来的过于依赖人力标注的不足。与传统的支持向量机模型相比,该模型成功地将准确率提高了4.92%。同时,该模型通过利用上下文语义信息,解决了词级别分类所带来的分类不准确问题。

With the rapid development of social networks and e-commerce,more and more users are accustomed to comment on products on the Internet,resulting in a rapid rise in the total number of product reviews on each major e-commerce site.In the face of the massive similar content and the format random comment,it is difficult for researchers and data users to extract valuable information from many short reviews by themselves,so the classification of emotions in short article comment has attracted wide attention.Aiming at the difficulty of manual extraction,we propose an improved model of convolution neural network.This model realizes the emotion classification of short text comments through the combination of word embedding and multi-channel convolution neural network,which makes up for the deficiency of too dependent on manpower annotation from the support vector machine model.Compared with the traditional support vector machine model,the proposed model successfully improves the accuracy by 4.92%.At the same time,it solves the problem of inaccurate classification caused by word-level classification through the use of contextual semantic information.