摘要
针对现有的虚假评论检测方法忽略了虚假评论文本的情感特征这一问题,提出一种基于注意力机制的多层编码器模型(ABME)。基于评论首尾部分表达情感更加强烈等特点,将评论拆分为首、中、尾三部分,提高首尾部分的权重;使用双向LSTM模型编码,得到三个局部表示,使用自注意力机制和注意力机制将三个局部表示编码成一个全局特征表示;通过Softmax分类器得到分类结果。实验结果表明,与目前的最好方法相比较,该模型的平均准确率提高了3.3%,平均精度提高了1.21%。
Aiming at the problem that the existing spam review detection methods neglect the emotional characteristics of spam reviews texts,we propose an attention-based multi-layer encoder model(ABME).Based on the characteristics of expressing more intense emotions at the beginning and the end of the review,the model divided the review into three parts:the first,the middle,and the last,and increased the weight of the first part and the last part.Afterwards,the bidirectional LSTM model was used to encode the three local representations.Then,three local representations were encoded into a global feature representation using the self-attention mechanism and attention mechanism.Finally,the classification results were obtained through the Softmax classifier.The experimental results show that compared with the current best method,the average accuracy of our model is improved by 3.3%and the average accuracy is improved by 1.21%.
作者
曾致远
卢晓勇
徐盛剑
陈木生
Zeng Zhiyuan;Lu Xiaoyong;Xu Shengjian;Chen Musheng(School of Software,Nanchang University,Nanchang 330047,Jiangxi,China;School of Software Engineering,Jiangxi University of Science and Technology,Nanchang 330013,Jiangxi,China)
出处
《计算机应用与软件》
北大核心
2020年第5期177-182,共6页
Computer Applications and Software
基金
国家社会科学基金项目(17BGL008)。
关键词
虚假评论检测
注意力机制
长短期记忆网络
表示学习
Spam review detection
Attention mechanism
Long-short term memory network
Representation learning
作者简介
曾致远,本科生,主研领域:自然语言处理,深度学习。;卢晓勇,教授。;徐盛剑,本科生。;陈木生,高工。
