基于cw2vec与BiLSTM的中文商品评论情感分类  

Sentiment Classification of Chinese Product Reviews Based on cw2vec and BiLSTM

作　　者：高统超 张云华 GAO Tong-chao;ZHANG Yun-hua(School of Information,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,浙江杭州310018

出　　处：《软件导刊》2020年第4期79-83,共5页Software Guide

摘　　要：针对方面级情感分类算法在中文领域商品评论中性能不佳的问题,从实际应用场景出发,基于cw2vec模型并结合BiLSTM模型,进行中文商品评论方面级情感分类。通过对数据进行预处理,训练中文词向量,提取评论语句文字笔画信息特征;然后对评论语料构建基于注意力机制的BiLSTM模型进行情感分类,计算注意力向量权重,利用双向网络结构特点捕捉语义依赖信息。实验结果表明,当训练语料分布合理时,该方法准确率达到83.2%,比Skip-gram模型提高了3.3%。该方法在中文方面级情感分类任务上能获取中文语义信息,分类效果更好,有效提高了分类准确率。Aiming at the poor performance of aspect level sentiment classification algorithm in Chinese commodity review,based on the actual application scenario,we combine the BiLSTM model with the cw2vec model to classify Chinese commodity reviews.The data is pre-processed,the Chinese word vector is trained,and the feature information of the comment sentence text is extracted.Then,the BiLSTM model based on the attention mechanism is constructed to classify the annotation corpus,calculate the weight of the attention vector,and capture the semantic dependence information by using the characteristics of the bidirectional network structure.The experimental results show that when the training corpus distribution is reasonable,the accuracy rate of this method is 83.2%,which is 3.3%higher than that of the Skip-gram model.This method can obtain Chinese semantic information on the Chinese aspect level sentiment classification task,and the classification effect is better,which effectively improves the classification accuracy.

关 键 词：情感分类 cw2vec模型 BiLSTM模型 注意力机制 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]