基于翻译模型的网络评论情感分析研究

发布时间：2019-01-21 12:10

【摘要】：随着电子商务的普及与发展,大部分网络消费者在购买商品前都会浏览商品的用户评论信息,这些评论信息不仅对消费者做购买决策有帮助,对生产厂家或电商公司获取商品市场反馈也有重要贡献。然而,从海量的商品评论中获取对用户有用的信息是一件很繁琐的事情,因此情感分析(Sentiment Analysis/Opinion Mining)应运而生。情感分析的主要工作就是挖掘并分析文本信息,实现相应的情感分类或情感极性判定。情感分析不仅在商品评论分析领域得到了广泛的应用,在网络舆情分析,金融股市分析和垃圾邮件处理等方面也有较高的研究价值。情感分析在自然语言处理,信息检索,数据挖掘以及机器学习等领域都有涉及,因此,研究情感分析的方法也比较全面。情感分析的主要内容分为两部分:倾向性信息识别和倾向性信息分类,在众多的研究方法中提高识别精度和分类精度一直是情感分析研究的重点。按分析的粒度的可以把情感分析分为篇章级别、句子级别、短语级别和词语级别。早期的情感分析主要集中在篇章级别和句子级别,随着用户提出的要求变高,短语级别和词语级别这种细粒度的情感分析逐渐成为了当前的研究热点。本文主要是对网络产品评论进行词语级别的情感分析研究,主要内容:1.研究了抓取网页数据的方法,并对数据进行了分词、词性标注和去噪声。2.研究识别情感词和评价对象词语的方法。情感分析中识别情感词和评价对象的精度一直是研究的重点,因此本文充分利用情感关系和词语本身的特性就如何有效地识别情感词和评价对象进行了详细的研究。除此之外,还实现了另外两种方法作为本文的对比实验,针对不同的抽取方法进行了详细分析。3.研究了情感词的情感倾向分析方法。把情感词的情感极性分为正面、中性和负面三个类别。首先分别用超限学习机、支持向量机和Softmax回归进行实验,然后选出分类结果做好的分类器。
[Abstract]:With the popularity and development of electronic commerce, most online consumers will browse the consumer comment information before purchasing the goods, which is not only helpful for consumers to make purchase decisions. For manufacturers or e-commerce companies to obtain feedback on the commodity market also has an important contribution. However, it is very tedious to get useful information from a large number of commodity reviews, so emotion analysis (Sentiment Analysis/Opinion Mining) emerges as the times require. The main work of affective analysis is to mine and analyze text information to realize the corresponding emotional classification or emotional polarity judgment. Affective analysis is not only widely used in the field of commodity comment analysis, but also has high research value in network public opinion analysis, financial stock market analysis and spam processing. Affective analysis is involved in natural language processing, information retrieval, data mining and machine learning. The main content of affective analysis is divided into two parts: the identification of tendentiousness information and the classification of tendentiousness information. Improving the recognition accuracy and classifying accuracy in many research methods has always been the focus of affective analysis research. Affective analysis can be classified into text level, sentence level, phrase level and word level according to the granularity of analysis. The early affective analysis mainly focuses on the text level and sentence level. As the demands of users become higher phrase level and word level such fine-grained emotional analysis has gradually become the focus of current research. This article is mainly to the network product comment carries on the word level emotion analysis research, the main content: 1. This paper studies the method of grabbing web page data, and carries on word segmentation, part of speech tagging and denoising. 2. 2. This paper studies the methods of identifying affective words and evaluating object words. The accuracy of identifying emotional words and evaluating objects in affective analysis has always been the focus of research. Therefore, this paper makes full use of the characteristics of affective relationships and words themselves to study in detail how to effectively identify affective words and evaluation objects. In addition, two other methods are implemented as the contrast experiment in this paper, and the different extraction methods are analyzed in detail. This paper studies the affective tendency analysis of affective words. The affective polarity of affective words is divided into three categories: positive, neutral and negative. First, the experiments are carried out by using the out-of-limit learning machine, support vector machine and Softmax regression respectively, and then the classifier with good classification results is selected.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】