期刊论文

Micro-blog Emotion Analysis Based on RAE ＋ Dropout

中文

情报工程

2095-915X

2017

3

6

44-53

10.3772/j.issn.2095-915x.2017.06.006

61402220:国家自然科学基金 61672178:国家自然科学基金 14YBA335:湖南省哲学社会科学基金 16YBA323:湖南省哲学社会科学基金 CX2016B446:湖南省研究生科研创新项目 LY13F020024:浙江省自然科学基金

本校为第一机构

文本情感分析是近年自然语言处理领域的研究热点之一,其中微博情感分析受到了学术界和企业界的广泛关注。微博情感分析是指对用户针对某一事件发表的言论进行正向、负向和中性情感的判定。本文在分析了标准RAE模型缺点的基础上,提出了一种基于RAE+Dropout的联合模型。该模型利用Dropout技术有效地预防过拟合问题的发生,同时也提高了模型训练速度。RAE+Dropout模型与RAE+词性选择模型、标准RAE模型以及SVM模型的对比实验结果表明:RAE+Dropout模型的准确率和F1值属于最优,比标准RAE模型的准确值和F1值高出0.82%和0.64%,尤其是在高维词语向量中RAE+Dropout模型的效果更加明显。摘要

Recently, text emotion analysis is one of the most popular research hotpots in the field of natural language processing, Furthermore, micro-blog emotion analysis has received extensive attention in academia and industry. Micro-blog emotion analysis aims to judge the positive, negative and neutral emotion of the speech delivered by the user for an event. In this paper, firstly, the shortcomings of the original model were illustrated according to the standard RAE model, and then an improved model, which is a joint model of RAE ＋ Dropout was put ...