基于深度学习的电商用户评论情感分类研究
Research on Sentiment Classification of E-Commerce User Reviews Based on Deep Learning
摘要: 电商平台中的用户评论不仅反映消费者对商品质量、物流服务和售后体验的真实感受,也是平台开展口碑监测、商品优化和风险预警的重要依据。由于中文电商评论普遍存在篇幅短、口语化表达多、评价维度分散以及噪声字符较多等特点,人工逐条分析往往效率低、成本高,难以满足实际业务需求。为提升电商评论情感识别的自动化水平,本文以Online Shopping 10 Cats数据集为基础,构建了融合BERT、BiGRU、Attention和FGM对抗训练的情感分类模型,并将其用于电商评论正负极性识别。实验结果表明,该模型在F1-score上达到0.9525,较对比模型具有一定优势。进一步分析发现,Attention机制有助于突出评论中的关键评价片段,FGM对抗训练能够增强模型对噪声表达的适应能力。研究结果说明,该方法可为电商平台评论监测、差评识别、商品口碑分析和用户反馈整理提供技术支持。
Abstract: User reviews on e-commerce platforms not only reflect consumers’ genuine feelings about product quality, logistics services and after-sales experience, but also serve as an important basis for platforms to conduct reputation monitoring, product optimisation and risk warning. Due to characteristics such as generally short length, colloquial expression, scattered evaluation dimensions and numerous noisy characters in Chinese e-commerce reviews, manual analysis of each review is often inefficient and costly, making it difficult to meet actual business needs. To improve the automation level of sentiment recognition in e-commerce reviews, this paper constructs a sentiment classification model based on the Online Shopping 10 Cats dataset, integrating BERT, BiGRU, Attention and FGM adversarial training, and applies it to recognising positive and negative sentiment in e-commerce reviews. Experimental results show that the model achieves an F1-score of 0.9525, demonstrating certain advantages over comparison models. Further analysis indicates that the Attention mechanism helps highlight key evaluation fragments within reviews, and FGM adversarial training can enhance the model’s adaptability to noisy expressions. The research results suggest that this method can provide technical support for review monitoring, negative review detection, product reputation analysis and user feedback management on e-commerce platforms.
参考文献
|
[1]
|
韩雨彤, 周季蕾, 任菲. 动态视角下实时评论内容对直播电商商品销量的影响[J]. 管理科学, 2022, 35(1): 17-28.
|
|
[2]
|
成耀, 张铎, 周宇, 等. 基于模糊聚类的电商企业不平衡财务数据风险预测方法[J]. 电子商务评论, 2025, 14(1): 640-647.
|
|
[3]
|
田依林, 刘璐, 莫洁安. 基于在线评论的跨境电商用户需求认知分析[J]. 情报科学, 2025, 43(11): 45-53.
|
|
[4]
|
徐文龙. 电商用户评论的半监督情感分类研究[D]: [硕士学位论文]. 北京: 北京交通大学, 2025.
|
|
[5]
|
王盈, 张文龙, 唐卓然. 基于电商评论情感分析的用户聚类方法研究[J]. 现代信息科技, 2023, 7(16): 24-27, 33.
|
|
[6]
|
高茂娇. 面向电商评论文本的跨领域情感分析研究[D]: [硕士学位论文]. 淮南: 安徽理工大学, 2024.
|
|
[7]
|
洪艺萍. 结合Bert和BiGRU的电商产品评论情感分析算法研究[J]. 佳木斯大学学报(自然科学版), 2026, 44(2): 91-94.
|
|
[8]
|
纪杨洋, 叶东琛. 跨平台评论信息的融合处理技术与聚类效果研究[J]. 情报理论与实践, 2025, 48(S1): 240-244.
|
|
[9]
|
何金凤, 沈玮, 张圣琪, 等. 基于改进BP神经网络的电商财务风险智能预测方法[J]. 电子商务评论, 2024, 13(4): 3022-3028.
|
|
[10]
|
朱子涵, 赵书娴, 程实. 人工智能时代老年数字阅读服务的适老化供给路径研究[J]. 老龄化研究, 2026, 13(9): 129-136. https://doi.org/10.12677/ar.2026.139635
|