面向网络商务系统评论的情感分析

发布时间：2018-05-11 08:41

本文选题：情感分析 + 电商评论　；参考：《东华大学》2017年硕士论文

【摘要】：随着互联网的发展特别是智能移动端的普及,越来越多的用户通过网络电商平台进行购物并留下评论。评论内容可以反映出用户的喜好以及商品的优缺点,直接影响着潜在客户的购买意向,也为生产厂商和销售企业提供决策依据。于是各种各样的情感分析技术被应用到评论领域。如何将这些评论内容的情感分析结果应用到生产实际中去是一个很值得研究的问题。评论中的句子不同于论坛中或者微博中的句子,它们大多数简短并且表达主语不明确,因此基于语义规则对评论进行研究能取得更好的效果,但是已有的语义规则并不能满足评论中特有的句子结构,有待进一步改进。本文首先介绍了情感分析的理论、背景、研究现状和相关技术;然后以服装电商评论为研究对象,提出一种基于语义规则的服装电商评论情感分析流程。重点总结了中文评论句子结构规则,提出了一种量化情感倾向强度的计算方法;进而提出了一种销量预测模型来研究情感倾向强度与销量之间的关系。通过使用分类算法分析影响销量的最佳评论页数;最后通过一个应用案例对研究成果进行验证。技术上主要采用Python语言实现,首先爬取淘宝上评分不同的两个商品4000多条评论数据建立原始语料库,并使用哈尔滨工业大学的语言技术平台对原始语料库进行分词和相关标注生成XML语料库;其次借助XML语料库和通用情感词典建立服装领域情感词典并使用该词典对XML语料库进行修正;然后采用本文提出的语义计算规则计算每条评论的情感倾向强度值;进而使用贝叶斯等七种分类算法分析情感倾向强度值、评论页数与销量之间的关系,使用召回率和精确度评估确定最佳评论页数;最后根据最佳评论页数分别使用线性回归模型、神经网络模型和支持向量机回归模型通过预测销量来分析情感倾向强度与销量之间的关系。
[Abstract]:With the development of the Internet, especially the popularity of intelligent mobile terminals, more and more users shop and leave comments through the network e-commerce platform. The content of comments can reflect the preferences of users and the advantages and disadvantages of products, directly affect the purchase intention of potential customers, and also provide decision basis for manufacturers and sales enterprises. As a result, a variety of emotional analysis techniques have been applied to the field of comment. How to apply the emotional analysis results of these comments to the production practice is a problem worth studying. The sentences in comments are different from those in forums or Weibo. Most of them are short and ambiguous, so it is more effective to study comments based on semantic rules. However, the existing semantic rules can not meet the specific sentence structure in comments, and need to be further improved. This paper first introduces the theory, background, research status and related technology of emotional analysis, and then takes the clothing e-commerce review as the research object, proposes a semantic rule based emotional analysis process of clothing ecommerce review. In this paper, the structure rules of Chinese comment sentences are summarized, a method to calculate the intensity of affective tendency is proposed, and a sales prediction model is proposed to study the relationship between the intensity of emotional tendency and the sales volume. The optimal number of comment pages affecting sales volume is analyzed by using the classification algorithm. Finally, an application case is used to verify the research results. Technically, it is mainly implemented in Python language. First of all, the original corpus is built by crawling more than 4000 comments on two items with different scores on Taobao. Using the language technology platform of Harbin University of Technology, the original corpus is segmented and annotated to generate XML corpus. Secondly, the XML corpus and the general emotion dictionary are used to establish the clothing domain emotion dictionary and the XML corpus is modified by the dictionary, and then the intensity of the emotional tendency of each comment is calculated by the semantic calculation rule proposed in this paper. Then seven classification algorithms, such as Bayes, are used to analyze the intensity of emotional tendency, the relationship between the number of comment pages and the sales volume, and the recall rate and accuracy evaluation are used to determine the best number of comment pages. Finally, linear regression model, neural network model and support vector machine regression model are used to analyze the relationship between emotional tendency intensity and sales volume by predicting sales volume.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F724.6;TP18

【相似文献】