基于ETW-BERT模型的网购商品虚假评论识别
Identification of false comments on online shopping products based on ETW-BERT model
陈润萌 1宋益善 2王胤哲 3梁靖韵4
作者信息
- 1. 华南师范大学软件学院,广州 510631
- 2. 香港中文大学(深圳)数据科学学院,深圳 518000
- 3. 华南农业大学数学与信息学院,广州 510642
- 4. 广州市第六中学,广州 510399
- 折叠
摘要
针对网购商品虚假评论识别问题,提出基于BERT双向预训练微调模型的假评识别方法.分析评论的文本、情感和时间特征,提出人工标注评论数据的12条规则,人工标注从京东网购平台爬取部分电子类产品的中文评论,获得5190条标注数据.对BERT的微调过程加入权重协方差对齐算法得到模型W-BERT,嵌入情感估值和时间特征得到模型ET-BERT,融合两者得到模型ETW-BERT.对上述标注数据集的实验表明,三个改进模型都取得了比BERT基础模型更好的效果.
Abstract
A false comment recognition method based on BERT bidirectional pre-training fine-tuning model is proposed to ad-dress the issue of false comment recognition for online shopping products.Analyze the text,emotional,and temporal characteristics of comments,propose 12 rules for manually annotating comment data.Manually annotating Chinese comments on some electronic products crawled from the JD online shopping platform,obtaining 5190 annotation data.The weight covariance alignment algorithm is added to the fine-tuning process of BERT to obtain the model W-BERT.The emotional estimation and temporal features are em-bedded to obtain the model ET-BERT,and the two are fused to obtain the model ETW-BERT.The experiments on the annotated da-taset mentioned above indicate that all three improved models have achieved better results than the BERT basic model.
关键词
虚假评论/BERT/预训练大模型/情感估值/时间特征Key words
false comments/BERT/pre-trained large models/emotional estimation/temporal features引用本文复制引用
出版年
2024
NETL
NSTL
万方数据