特征加权融合的朴素贝叶斯情感分类算法  

Naive Bayesian algorithm for text sentiment classification based feature weighting integration

作　　者：曾宇[1,2] 刘培玉[1,2] 刘文锋[1,3] 朱振方[4] 

机构地区：[1]山东师范大学信息科学与工程学院,山东济南250358 [2]山东省分布式计算机软件新技术重点实验室,山东济南250358 [3]菏泽学院计算机与信息工程系,山东菏泽274015 [4]山东交通学院信息科学与电气工程学院,山东济南250357

出　　处：《西北师范大学学报（自然科学版）》2017年第4期56-60,73,共6页Journal of Northwest Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61373148;61502151);教育部人文社科项目(14YJC860042);山东省自然基金资助项目(ZR2014FL010)

摘　　要：为解决文本情感分类准确率不高的问题,提出了一种特征加权融合的朴素贝叶斯情感分类算法.通过分析单个情感词对文本情感分类的贡献度特征,根据情感词对文本情感贡献度的权值调整贝叶斯模型的后验概率;将文本中所有相同极性的情感词作为一个特征整体,根据特征整体对文本情感贡献度的权值调整贝叶斯模型的整体概率.为了进一步提高分类的准确率以及提升分类模型的综合性能,将两种加权方式同时与朴素贝叶斯模型结合.结果表明,融合后的方法在数据集上的整体平均查准率、查全率分别提高1.83%和3.42%,平均F1值提高了2.76%.In order to improve the accuracy rate of text sentiment classification ,a naive Bayesian algorithm for text sentiment classification based feature weighting integration is proposed . Firstly , by analyzing the feature of the individual sentiment word contribute to the text sentiment classification , it adjusts the posteriori probability of the Bayesian model according to the weight value of the sentiment words ＇contribution to the text sentiment classification . Secondly , all sentiment words of same polarity are treated as a whole whose feature is merged with Bayesian model and the probability of the Bayesian model that is adjusted according to the weight value of the feature＇s contribution . Finally , to improve the accuracy rate and enhance the comprehensive performance of the classification model , the two weighting methods are integrated into Bayesian model . The experimental results illustrate that the overall average precision and the recall of the integrated method on the dataset are increased by about 1.83% and 3.42% respectively ,and the average F1 value increases by about 2.76% .

关 键 词：情感分类 特征加权 贡献度特征 朴素贝叶斯 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]