网络评论文本的情感倾向性研究

发布时间：2018-03-17 12:53

本文选题：情感分析　切入点：情感词典　出处：《暨南大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着近几年来电子商务的快速发展,越来越多的用户在网上购买商品后会留下对商品或服务的评论,并且该类评论数据的数量正在不断增长。这些数据不仅可以作为消费者购物的参考信息,也可以为商家的销售策略提供很多建议和商机,因此研究这些评论文本数据具有很大价值。作为文本分析的重要技术之一,文本情感分析在近几年得到了研究人员的广泛关注,并取得了很多的研究成果。文本情感分析根据评论文本中观点的褒贬倾向来进行文本分类,其分类方法包括基于情感词典的方法和基于机器学习的方法,其中第一种方法易于实现但是领域针对性不强,第二种方法分类效果更好但是需要大量的训练语料。本文首先介绍了情感分析的研究背景、国内外研究现状以及关键技术。然后对比了基于情感词典和基于机器学习两种方法的优缺点,并且结合机器学习中深度模型的有效性,选取了长短型记忆递归网络作为分类模型,提出将情感词典与深度学习模型两者结合的新方法来分阶段进行文本分类。新方法既克服了情感词典的不完整性缺点也解决了机器学习的大量语料需求。最后,本文通过对比实验证明在缺少训练语料情况下,将情感词典和长短型记忆递归网络结合的方法应用在文本分类中可以获得较高的精确度。
[Abstract]:With the rapid development of electronic commerce in recent years, more and more users will leave comments on goods or services after buying goods on the Internet. And the number of such reviews is growing. These data not only serve as a reference for consumers to shop for, but also provide a lot of advice and business opportunities for businesses' sales strategies. As one of the important techniques of text analysis, text emotion analysis has been paid much attention by researchers in recent years. A lot of research results have been made. Text affective analysis carries out text categorization according to the positive or negative tendency of opinion in the comment text. The classification methods include affective dictionary based approach and machine learning based approach. The first method is easy to implement but the domain is not specific. The second method has better classification effect but requires a lot of training data. Firstly, this paper introduces the research background of emotion analysis. Then, the advantages and disadvantages of affective dictionary and machine based learning are compared, and the validity of depth model in machine learning is combined. The long-long memory recursive network is selected as the classification model. A new method, which combines emotion dictionary with depth learning model, is proposed to classify text in stages. The new method not only overcomes the incompleteness of emotion dictionary, but also solves the need of a large amount of corpus for machine learning. In this paper, it is proved that in the absence of training corpus, the combination of affective dictionary and long-memory recursive network can achieve high accuracy in text classification.
【学位授予单位】：暨南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】