吕妹园 张永健 张永强 孙胜娟

摘 要：為了解决分布式词表示方法因忽略词语情感信息导致情感分类准确率较低的问题，提出了一种融入情感信息加权词向量的情感分析改进方法。依据专属领域情感词典构建方法，结合词典和语义规则，将情感信息融入到TF-IDF算法中，利用Word2vec模型得到加权词向量表示方法，并运用此方法对采集到的河北省旅游景点的评论文本与对照组进行对比实验。结果表明，与基于分布式词向量表示的情感分析方法相比，采用融入情感信息加权词向量的改进方法进行情感分析，积极文本的准确率提高了6.1%，召回率提高了6.6%，F值达到了90.3%;消极评论文本的准确率提高了6.0%，召回率提高了7.2%，F值达到了89.6%。因此，融入情感信息加权词向量的情感分析改进方法可以有效提高评论文本情感分析的准确率，为用户获得更为准确的评论观点提供参考。

关键词：自然语言处理;语义规则;情感信息;TF-IDF;Word2vec;加权词向量;情感分析

中图分类号：TP391.1   文献标识码：A

doi：10.7535/hbkd.2021yx04008

收稿日期：2021-03-25;修回日期：2021-06-11;责任编辑：王淑霞

基金项目：河北省创新能力提升计划项目（19456003D）

第一作者简介：吕妹园（1996—），女，山东济南人，硕士研究生，主要从事自然语言处理方面的研究。

通讯作者：张永强教授。E-mail：120030009@qq.com

吕妹园，张永健，张永强，等.融入情感信息词向量的评论文本情感分析方法[J].河北科技大学学报，2021，42（4）：380-388.LYU Meiyuan，ZHANG Yongjian，ZHANG Yongqiang， et al.Sentiment analysis method of comment text based on word vector with sentiment information[J].Journal of Hebei University of Science and Technology，2021，42（4）：380-388.

Sentiment analysis method of comment text based on word vector with sentiment information

LYU Meiyuan，ZHANG Yongjian，ZHANG Yongqiang，SUN Shengjuan

（School of Information and Electrical Engineering，Hebei University of Engineering，Handan，Hebei 056107，China）

Abstract：In order to solve the problem of low accuracy of sentiment classification caused by neglecting the sentiment information of words in distributed word representation method，an improved sentiment analysis method incorporating weighted word vectors of sentiment information was proposed.According to the exclusive domain sentiment dictionary，combined with the dictionary and semantic rules，the sentiment information is integrated into the TF-IDF algorithm，and the weighted word vector representation method is obtained by using word2vec model.The method is used to compare the collected comments of tourist attractions in Hebei Province with the control group.The results show that compared with the sentiment analysis method based on distributed word vector representation，the accuracy and recall rate of positive text are increased by 6.1% and 6.6%，and the Fvalue reached 90.3%，the accuracy and recall rate of negative text are increased by 6.0% and 7.2%，and the Fvalue reached 89.6% by using the improved method of sentiment analysis integrated with sentiment information weighted word vector.Therefore，the improved method of sentiment analysis integrated with sentiment information weighted word vector can effectively improve the accuracy of sentiment analysis of comment text，and provide valuable reference for users to obtain more accurate comments.

Keywords：

natural language processing;semantic rules;sentiment information;TF-IDF;Word2vec;weighted word vector;sentiment analysis

随着互联网的发展，越来越多的互联网用户开始在线上发表自己的观点，如淘宝、携程网等平台上用户对商品和景点的评论，情感分析技术可以让用户更便捷地获取评论的情感倾向。……