采用情感特征向量的Twitter情感分类方法研究

摘要

面向公共媒体内容开展情感分析是分析公众情感的一项基础工作.经典的基于词频特征向量的特征提取方法,主要利用词频作为文本分类的依据,而词频与情感信息之间的关系并不紧密.提出一种采用基于情感特征向量的Twitter推文情感分类方法.该方法首先通过对推文进行数据清洗、词形还原、词性标注和词汇向量化;其次,将单词匹配到情感词典中;最后,利用每个单词的正向情感、负向情感取值生成情感特征向量,通过MNB、SVM等机器学习方法训练模型,对推文的情感进行分类.实验结果表明采用情感特征向量的Twitter推文情感分类方法能够获得更佳的分类性能.%Sentiment classification on the content of the public media is a fundamental task for analyzing public sentiment.A classic model for sentiment classification is based on word frequency model which takes advantages of word frequency in text classification.However,the relationship between word frequency and sentiment information is not actually close as we expected.This paper presents a distinct approach of sentiment classification using sentimental feature vector instead of word-frequency feature vector.First,preprocessing is done to clean,lemmatize,and POS tag each word in a single tweet;Second,with the sentiment dictionary,each word is attached with a score corresponding to positive or negative sentiment respectively so as to get the sentimental feature vector for each tweet;Third,sentiment of tweets are classified by training models of different algorithms such as Multinomial Na? ve Bayes (MNB) and Support Vector Machine (SVM).Empirical studies show that our sentimental feature vector is beneficial for Twitter sentiment classification.