融合多特征的在线评论情感分类

Online Review Sentiment Classification Incorporating Multiple Features

多数基于深度学习的情感分类方法过于追求数据驱动,忽视了文本情感特征对于情感分类的影响;而一些融合情感特征的情感分类方法仅利用了部分相关特征,使得其他情感特征并没有得到充分利用.针对这一现象,提出了一种融合词级特征和句级特征的在线评论情感分类模型.首先利用构建的情感元素词典获取情感词、否定词等多种特征词,然后通过多维特征向量表示将多种文本特征转化为词级特征向量和句级特征向量,最后将这些特征向量融入LSTM网络模型完成情感分类,整个模型简称为MF-LSTM(sentiment classification model based on multidimensional features and LSTM).MF-LSTM充分利用了评论文本的情感先验知识,情感分类能力得到显著提升.在3个中文评论数据集上的实验结果表明MF-LSTM模型相比其他深度学习方法具有更好的分类效果,并且在样本数据不平衡的情况下也能具有较好的鲁棒性.

Most deep learning-based sentiment classification methods are too data-driven and ignore the impact of text sentiment features on sentiment classification;while some sentiment classification methods that fuse sentiment features only utilize some relevant features,making other sentiment features not fully utilized.To address this phenomenon,an online comment sentiment classification model based on the fusion of word-level features and sentence-level features is proposed.Firstly,the constructed sentiment element dictionary is used to obtain various feature word such as sentiment words and negation words.Then multiple text features are transformed into word-level feature vectors and sentence-level feature vectors through multidimensional feature vector representation.Finally,these feature vectors are integrated into the LSTM network model to complete sentiment classification.The whole model is called MF-LSTM(sentiment classification model based on multidimensional features and LSTM).MF-LSTM makes full use of the sentiment prior knowledge of the review text and significantly improves the sentiment classification ability.The experimental results on three Chinese commentary datasets show that the MF-LSTM model has better classification results than other deep learning methods and is more robust in the case of unbalanced sample data.