基于文本情感分析技术的用户评论分析系统设计与实现

发布时间：2018-03-22 20:23

本文选题：情感分析　切入点：用户评论　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：互联网技术的不断发展促进了电子商务发展,网络购物与人们的生活越来越密切,这种情况下便产生了大量针对各种商品和服务的用户评论数据,这些评论对用户和商家来讲都具有很重要的利用价值。但是互联网信息变化很大很快,网络上用户评论信息数据量很大,同时也具有很强的随意性,这导致大量的信息中有很多垃圾信息。如果通过人工的方式去处理会耗费大量的人力物力成本,效果也不理想。因此需要计算机来帮助用户和商家获取信息并进行分析,文本情感分析技术则能够很好的处理这项任务。本文以文本情感分析技术理论为指导,通过对产品评论的情感分析方法进行研究,设计出一套用户评论情感分析系统,该系统能够根据用户的需求自动的获取网上的信息,并对网上的垃圾信息进行过滤,然后使用情感分析技术对这些评论数据进行分析处理,自动的识别出用户对相关产品以及产品属性的喜好程度,归纳出商品的正负向评论信息以及涉及到的产品属性。最终通过可视化技术将结果展示给用户和企业。主要研究内容如下:(1)在双向传播算法的基础上提出一种基于规则和基于统计相结合的产品属性和用户观点抽取方法。根据中文语言词语的特殊性,将中文语言中的依存句法关系这种规则与双向传播这种基于统计的算法结合起来,从而能够提高抽取产品属性和用户观点词的准确率,然后通过点互信息法对抽取信息进行过滤,将冗余信息去除掉,最后形成产品属性和用户观点词的对应关系;(2)设计一个改进的词典法,将机器学习中的支持向量机算法引入。首先手工制作出一套情感词典,使用词典法对整个文本的情感值进行计算,这一步对整个文本进行初步的筛选,然后利用支持向量机对文本进行分类处理,经过实验分析将支持向量机引入情感极性识别中效果比较好。(3)设计出一套可视化展示方案,将文本情感分析的结果通过信息可视化的方式展示出来,根据不同的结果选择不同的图形进行展示。并对效果进行评估调查。(4)最后对整个系统的效果进行评测,并结合可视化设计的效果进行用户体验方面的满意度调查,制作调查问卷,通过对调查问卷数据进行分析总结。
[Abstract]:The continuous development of Internet technology has promoted the development of electronic commerce, and online shopping is becoming more and more close to people's life. In this case, a large number of user comments on a variety of goods and services have been generated. These comments are of great value to users and businesses. But the information on the Internet changes very quickly, and the amount of data that users comment on on the Internet is very large, and at the same time, it has a strong arbitrariness. This leads to a lot of junk information in a large amount of information. If it is handled manually, it will cost a lot of manpower and material resources, and the effect will not be satisfactory. Therefore, computers are needed to help users and businesses to obtain information and analyze it. The text emotion analysis technology can deal with this task very well. Under the guidance of the text emotion analysis technology theory, this paper designs a set of user comment emotion analysis system through the research on the emotion analysis method of product comment. The system can automatically obtain the information of the network according to the needs of the users, filter the garbage information on the network, and then use the emotion analysis technology to analyze and process these comments data. Automatically identify the user's preference for related products and product attributes, The positive and negative comment information of goods and the product attributes involved are summarized. Finally, the results are presented to users and enterprises through visualization technology. The main research contents are as follows: 1) A basis based on bidirectional propagation algorithm is proposed. Product attribute and user viewpoint extraction method based on rules and statistics. According to the particularity of Chinese language words, By combining the rule of dependency syntax in Chinese language with the statistical algorithm of bidirectional propagation, the accuracy of extracting product attributes and user view words can be improved. Then filter the extracted information through the point mutual information method, remove the redundant information, and finally form the corresponding relationship between product attributes and user opinion words. (2) Design an improved dictionary method. The support vector machine (SVM) algorithm in machine learning is introduced. Firstly, a set of emotion dictionary is made by hand, and the emotion value of the whole text is calculated by using the dictionary method. Then we use support vector machine to classify the text and introduce support vector machine into emotional polarity recognition through experimental analysis. A visual display scheme is designed. The result of text emotion analysis is displayed through information visualization, and different graphics are selected according to different results. Finally, the effect of the whole system is evaluated. Combined with the effect of visual design, the satisfaction survey of user experience is carried out, the questionnaire is made, and the data of the questionnaire is analyzed and summarized.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.52;TP391.1

【参考文献】