基于多分类器投票集成的中文评论情感分类  

Chinese Comment Emotional Classification Based On Multi-classifier Voting Rule

作　　者：苗鹏宇 刘征宏 代宣军[1] MIAO Peng-yu;LIU Zheng-hong;DAI Xuan-jun(College of Mechanical and Control Engineering,Guilin University of Technology,Guilin Guangxi 541006,China;College of Mechanical Engineering,Guiyang University,Guiyang Guizhou 550005,China)

机构地区：[1]桂林理工大学机械与控制工程学院,广西桂林541006 [2]贵阳学院机械工程学院,贵州贵阳550005

出　　处：《计算机仿真》2025年第2期494-500,共7页Computer Simulation

基　　金：国家自然科学基金资助项目(52105248);贵州省教育厅科技拔尖人才项目(黔教技[2022]086);贵阳市科技计划项目([2023]48-18)。

摘　　要：情感分类是指对网络在线评论进行自动极性分类,如正极、负极、中性。针对单一情感分类模型可能存在不稳定的问题,研究多分类器系统概念在中文评论情感分类的潜在益处,并提出一种基于集成学习的情感分类方法。首先改进朴素贝叶斯(NB)为自适应朴素贝叶斯,对支持向量机(SVM)进行参数优化,同时以SVM为基分类器将元分类器装袋(Bagging)作为集成组件之一,然后使用投票集成组合三个组件并输出分类结果,最后将投票集成与各基分类器、元分类器单独使用在三个数据集上的表现进行对比。实验表明,基于绝对多数投票法的集成分类在股市评论、网购评论、电影评论三个数据集上准确率(Accuracy)、精确率(Precision)、召回率(Recall)、F值(F-measure)均有所提升。由此可见,提出的方法有效地提高了中文评论文本情感分类性能。Sentiment classification refers to the automatic polarity classification of online comments,such as positive,negative,and neutral.Aiming at the possible instability of a single sentiment classification model,the potential benefits of the multi-classifier system concept in Chinese review sentiment classification are studied,and an ensemble learning-based sentiment classification method is proposed.Firstly,Naive Bayesian(NB)is improved to Adaptive Naive Bayesian(Ada_NB),parameter optimization is performed on Support Vector Machine(SVM),and meta-classifier Bagging is used as one of the integrated components,and then the voting ensemble is used to combine the three components and outputs the classification results,and finally,the performance of the voting ensemble is compared with that of each base classifier and meta-classifier when used individually on the three datasets.Experiments have shown that ensemble classification based on the absolute majority voting method has improved accuracy,precision,recall,and F-measure on three datasets:stock market reviews,online shopping reviews,and movie reviews.It can be seen that the method proposed in this paper effectively improves the sentiment classification performance of Chinese comment texts.

关 键 词：情感分类 集成学习 机器学习 中文评论 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]