基于双层注意力机制的用户创新评论提取方法  

User Innovation Review Extraction Method Based on Two-layer Attention Mechanism

作　　者：于涛 王瑞 战洪飞[1] 余军合[1] YU Tao;WANG Rui;ZHAN Hongfei;YU Junhe(School of Mechanical Engineering and Mechanics,Ningbo University,Ningbo Zhejiang 315000,China)

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315000

出　　处：《机械设计与研究》2024年第3期7-12,18,共7页Machine Design And Research

基　　金：浙江省省属高校基本科研业务费项目(SJLZ2023001);国家重点研发计划资助项目(2019YFB1707101,2019YFB1707103);国家自然科学基金资助项目(71671097)。

摘　　要：电商平台积累着大量用户发表的商品评论,蕴含着丰富的产品创新想法,可以为产品设计人员提供创意与灵感。然而,传统的评论挖掘技术在识别语义复杂且不规范的创新评论时,精度较低,无法满足实际需求。为此,我们提出一种基于双层注意力机制的双层BiLSTM模型,以高精度地提取创新评论。该模型首先基于预测策略的EDA数据增强方法扩充创新评论,然后在Bert预训练模型的基础上结合词向量注意力机制生成评论句子中单词的语义和上下文表示,再利用双层BiLSTM提取评论语句的时序特征,最后通过注意力机制识别评论语句的关键特征。通过对两个商品评论数据集进行实例验证,表明本文提出的方法比目前的创新评论提取方法识别精度更高,F1值可以达到92%。E-commerce platforms have accumulated a large number of product reviews published by users,which contain rich product innovation ideas that can provide creativity and inspiration for product designers.However,traditional review mining techniques have low accuracy when identifying innovative reviews with complex semantics and irregularities,which cannot meet the actual needs.To this end,we propose a two-layer BiLSTM model based on a two-layer attention mechanism to extract innovation reviews with high accuracy.The model first uses a screening EDA data augmentation technology to expand innovative reviews,and then combines the word vector attention mechanism on the basis of Bert pre-training model to generate word semantics and context representation information,and then uses two-layer BiLSTM to extract the timing features of review sentences.Finally,the attention mechanism is used to identify the key features of review sentences,so as to accurately identify innovation reviews.Through the example verification of two product review data sets,it shows that the method proposed in this paper has higher recognition accuracy than the current innovation review extraction method,and the F1 value can reach 92%.

关 键 词：信息提取 深度学习 文本分类 电商平台 在线评论 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]