基于加权集成的序贯三支决策情感分类  

Sentiment Classification Using Sequential Three-way Decisions Based on Weighted Ensemble

作　　者：帅常朗 钱进 周川鹏 SHUAI Changlang;QIAN Jin;ZHOU Chuanpeng(School of Software,East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]华东交通大学软件学院,江西南昌330013

出　　处：《山西大学学报(自然科学版)》2025年第1期66-76,共11页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(62066014,62466017);江西省自然科学基金项目(20232ACB202013)。

摘　　要：为了提高情感分类的性能,本文提出了一种基于加权集成的序贯三支决策情感分类模型。该模型首先对评论数据集的边界域使用不同的分类器获取各自的预测概率,再根据历史分类性能对不同分类器的预测概率进行加权集成,然后根据阈值和代价损失分别进行三支决策,将评论划分为正类、负类和边界域。对于边界域进行序贯的集成概率预测,并根据概率和阈值进一步划分为新的正类、负类和边界域。直至最细粒度上的边界域,最终通过集成二支决策得到最终的分类结果。研究结果表明,该模型在中文计算机评论、酒店评论和服装评论数据集上性能优于现有方法,其中在酒店评论数据集上分类准确率达到86.75%,相比于基于硬投票集成的序贯三支决策情感分类提高了3.6%。To improve the performance of sentiment classification,this paper proposes a sentiment classification model based on weighted ensemble sequential three-way decision.The model firstly uses different classifiers to obtain their respective prediction probabilities for the boundary domain of the review dataset.Then,based on historical classification performance,the prediction probabilities of different classifiers are weighted and integrated.According to the threshold and cost loss,three-way decisions are made to classify the reviews into positive,negative,and boundary domains.The boundary domain is sequentially subjected to integrated probability prediction and further classified into new positive,negative,and boundary domains according to the probabilities and thresholds.Until the finest granularity of the boundary domain is reached,a final classification result is obtained through integrated two-way decision.The research results show that this model outperforms existing methods on Chinese computer reviews,hotel reviews and clothing reviews datasets.Among them,the classification accuracy on the hotel review data set reached 86.75%,which was improved by 3.6%compared with the sequential three-branch decision emotion classification based on hard voting integration.

关 键 词：多粒度分类 机器学习 集成学习 文本粒化 粗糙集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]