基于LSTM的商品评论文本情感分类研究

Research on Sentiment Classification of Commodity Reviews Based on LSTM

近年来,人们的消费方式逐渐由线下转变为线上,在线评论文本在人们是否进行消费活动中起到了主导作用,但不同的人对商品的不同方面的重视程度不同,所以笔者对传统的单一的文本情感分析方法进行了改进。首先根据评价对象对评论进行分类,将经过预处理的数据输入LDA模型中进行主题分类,得到文档的主题概率分布,将Word2vec模型运用到LDA模型中,获得主题扩展特征,将其与文本词向量相结合,获得融合后的特征向量,然后利用LSTM模型进行情感分类。实验证明,此模型能有效改善传统情感分类方法无法根据评价对象进行分类的弊端,有效提高了评论文本分类的性能。

In recent years,people’s consumption mode has gradually changed from offline to online.Online comment text plays a leading role in whether people conduct consumption activities,but different people attach different importance to different aspects of goods,so the author improves the traditional single text emotion analysis method.Firstly,the comments are classified according to the evaluation object,the preprocessed data is input into LDA model to classify the topics,and the topic probability distribution of the document is obtained.Word2vec model is applied to LDA model to obtain the extended features of the topics,which are combined with the text word vector to obtain the fused feature vector,and then LSTM model is used to classify the emotions.Experiments show that this model can effectively improve the shortcomings of traditional emotion classification methods that can not be classified according to the evaluation object,and effectively improve the performance of comment text classification.