摘要
为了进一步提高电商评论数据情感分类的准确率,提出了融合BERT词嵌入与多通道CNN(BERT-MCNN)的电商评论情感分类模型。模型利用BERT的多头注意力机制从不同角度对评论信息进行语义表达;再利用改进型多通道CNN模型(MCNN)的多通道卷积操作与多通道池化操作增强特征之间的关联性,使得模型更好地获得评论中的有用特征。在公开数据集SP以及自建的京东苹果手机评论数据集上进行性能验证实验,BERT-MCNN模型均获得了最高的94.51%和98.32%的准确率。通过对两组实验结果分析,得出在短评论情感分析中MCNN的分类效果胜于CNN,LSTM与RNN的结论,证明了所提方法的对电商评论文本自动划分情感类别的可行性和有效性。
To further enhance the accuracy of sentiment classification for e-commerce reviews,a novel sentiment classification model integrating BERT word embeddings and Multi-Channel CNN(BERT-MCNN)is proposed.The model leverages BERT's multi-head attention mechanism to capture semantic representations of review information from diverse perspectives.Subsequently,the enhanced Multi-Channel CNN(MCNN)architecture employs multi-channel convolutional operations and multi-channel pooling operations to strengthen feature correlations,enabling more effective extraction of useful features from reviews.Experimental validation was conducted on both the publicly available dataset SP and a self-built JD.com Apple smartphone review dataset.The BERT-MCNN model achieved the highest accuracy rates of 94.51%and 98.32%on these datasets,respectively.Analysis of the experimental results demonstrates that MCNN outperforms conventional CNN,LSTM,and RNN architectures in short review sentiment analysis,confirming the feasibility and effectiveness of the proposed method for automatic sentiment categorization of e-commerce review texts.
作者
黄萍
朱惠娟
郑磊
李菊
HUANG Ping;ZHU Huijuan;ZHENG Lei;LI Ju(College of Computer and Artificial Intelligence,Nanjing University of Science and Technology ZiJin College,Nanjing 210023,China)
出处
《佳木斯大学学报(自然科学版)》
2025年第12期21-24,共4页
Journal of Jiamusi University:Natural Science Edition
基金
江苏省高校哲学社会科学研究项目(2023SJYB0677)
南京理工大学紫金学院校级科研项目(2023ZRKX0401007)。
关键词
BERT
注意力机制
卷积神经网络
多通道卷积神经网络
电商评论分类
情感分类
BERT
attention mechanism
convolutional neural network
multi-bandwidth convolutional neural network
e-commerce review classification
sentiment classification
