基于陶瓷产品评论的情感分析

Sentiment Analysis Based on Ceramic Product Reviews

摘要: 本文基于陶瓷产品的评论数据进行了评论文本情感分析，本文主要研究的是应用机器学习方法对陶瓷产品评论文本进行情感分类，采用的机器学习情感分类模型有：朴素贝叶斯(NB)、支持向量机(SVM)、人工神经网络(ANN)和随机森林(RF)。在情感分类后进行了对比分析，对比分析后发现：1) 在此数据集下，Word2vec词向量文本特征表示方法相比于词频矩阵和TF-IDF这两种文本特征表示方法更适合于文本特征表示；2) 在此数据集下，四种情感分类模型中，人工神经网络(ANN)的分类效果最差，支持向量机(SVM)的分类效果最好。综上所述，对于陶瓷产品评论文本情感分类，在此数据集下，这三种文本特征表示方法和四种机器学习情感分类模型的所有组合中，Word2vec词向量方法和支持向量机的组合的情感分类效果最好。

Abstract: In this paper, sentiment analysis of review texts is conducted based on the review data of ceramic products. The main research of this paper is to apply machine learning methods to classify the sentiment of ceramic product review texts, and the machine learning sentiment classification models used are: plain Bayesian (NB), support vector machine (SVM), artificial neural network (ANN) and random forest (RF). A comparative analysis was conducted after sentiment classification, and the comparative analysis revealed that: 1) under this dataset, the Word2vec word vector text feature representation method is more suitable for text feature representation than the two text feature representation methods of word frequency matrix and TF-IDF; 2) under this dataset, among the four sentiment classification models, artificial neural network (ANN) has the worst classification effect and support vector machine (SVM) has the best classification effect. In summary, for the sentiment classification of ceramic product review texts, among all combinations of these three text feature representation methods and four machine learning sentiment classification models in this dataset, the combination of Word2vec word vector method and support vector machine has the best sentiment classification effect.