基于多层注意力机制的服装电商评论情感分析-《计算机技术与发展》

文章信息/Info

Title:: Sentiment Analysis of Clothing E-Commerce ReviewsBased on BiGRU-SD-Attention

作者:: 胡新荣1 ; 2 ; 王哲1; 2 ; 刘军平1; 2* ; 彭涛1; 2 ; 何儒汉1; 2; 1. 湖北省服装信息化工程技术研究中心,湖北武汉 430200;
2. 武汉纺织大学数学与计算机学院,湖北武汉 430200

Author(s):: HU Xin-rong1; 2 ; WANG Zhe1; 2 ; LIU Jun-ping1; 2 * ; PENG Tao1; 2 ; HE Ru-han1; 2; 1. Engineering Research Center of Hubei Province for Clothing Information,Wuhan 430200,China;
2. School of Mathematics and Computer,Wuhan Textile University,Wuhan 430200,China

Keywords:: distributed crawler; clothing e-commerce review; bidirectional gate recurrent unit; attention mechanism; sentiment analysis

摘要:: 互联网购物逐渐走进人们生活,人们在购物的同时也会留下海量评论文本,这些文本蕴含着巨大的价值和情感倾向,通过分析这些服装电商评论文本情感倾向,为推荐系统提供了参考。传统的算法难以提取到文本更深层次的情感特征,难以达到很好的效果。因此,该文提出了一种基于多层注意力机制 BiGRU-SD-Attention 的算法模型。首先,通过分布式爬虫采集服装电商评论文本,将文本数据进行清洗,划分为词语级别和句子级别数据集;利用 BiGRU 网络提取文本的正、负情感特征,然后对词语和句子分别运用注意力机制进行情感特征的重新加权计算;通过多层递进的权重计算,最后分类输出服装电商文本的情感特征倾向。实验结果表明,该算法的准确率达到了 94. 23% , 对比传统的 SVM 算法(81郾 67% ) 以及单一注意力机制的 BiLSM-Attention 算法(93. 50% ) ,在各方面都有了显著的提升。

Abstract:: Internet shopping has gradually come into people’s life. People will leave a large number of comments while shopping,whichcontain great value and emotional tendency. By analyzing the emotional tendency of these comments,it also provides reference value forthe recommendation system. The traditional algorithm is difficult to extract the deeper emotional features of the text,so it is difficult to achieve better results. Therefore,we propose an algorithm model based on the multi-layer attention mechanism BiGRU-SD-Attention.Firstly,the apparel e-commerce comment text is collected by distributed crawler,and the text data is cleaned and divided into word-leveland sentence-level data sets. Secondly,BiGRU network is used to extract the positive and negative emotional features of text,and thenthe attention mechanism is used to re-weight the emotional features of words and sentences respectively. Through multi-layer progressiveweight calculation,the emotional characteristic tendency of apparel e-commerce text is finally classified and output. Experiment showsthat the accuracy of the proposed algorithm reaches 94. 23% ,which is significantly improved in all aspects compared with the traditionalSVM algorithm (81. 67% ) and BiLSM-Attention algorithm with single attention mechanism (93. 50% ) .