哈尔滨工业大学管理学硕士学位论文
- I -
摘要
随着电子商务的普及，在线评论的研究经历了一个由易到难的过程，近
几年，由于文本挖掘技术的提高，越来越多的学者把目光转向文本评论。但
是，目前文本评论对销量的影响研究存在相互矛盾的结论，这很可能是由于
不合理的文本分类所致。文本分类依赖于准确的情感分析方法，但目前中文
情感分析方法有待进一步提高
本文利用大数据挖掘工具Python抓取109个平板电脑的1227861条数
据，根据最佳唤醒理论、稀释效应、从众效应、说服与知晓效应等理论，以
839份消费者在线评论浏览路径调查问卷调查结果为参考，利用动态Panel模
型对数据进行回归。为提高文本分类的准确性，本文从分词工具的选择、词
库的扩充以及文本评论的分类等三方面改进了文本挖掘技术。通过分词工具
的选择与词库的扩充，提升了文本赋值的准确率；通过二维文本分类方法，
结合消费者网络购物实践，扩充了中性文本评论的取值范围，提高了文本分
类的科学性与合理性；通过大数据挖掘工具Python，将在线文本评论细分为
积极文本评论、混合偏积极文本评论、混合中性文本评论、无差异中性文本
评论、混合偏消极文本评论与消极文本评论六个子类。其中，积极性质的文
本评论包括积极文本评论与混合偏积极文本评论，回归结果表明，积极文本
评论与产品销量呈现“U”型关系；混合偏积极文本评论与产品销量呈正相关
关系。中性文本评论包括混合中性文本评论与无差异中性文本评论，回归结
果表明，混合中性文本评论与产品销量呈正相关关系；无差异中性文本评论
与产品销量呈负相关关系。消极性质的文本评论包括混合偏消极文本评论与
消极文本评论，回归结果表明，混合偏消极文本评论与产品销量呈正相关关
系；消极文本评论与产品销量呈负相关关系。关于评论数量对文本评论与产
品销量之间的关系的调节作用，结果表明，评论数量的调节作用表现为提高
产品销量。其中，评论数量对积极文本评论、混合中性文本评论以及混合偏
消极文本评论与产品销量之间的关系起到正向调节作用，对无差异中性文本
评论与产品销量之间的关系起到负向调节作用。研究结果对解释目前在线文
本评论研究领域存在相互矛盾研究结论的现象具有一定的参考价值，同时对
电商平台维护在线评论、促进产品销量有一定的指导作用
关键词：文本评论；情感分析；二维文本分类；情感字典
哈尔滨工业大学管理学硕士学位论文
- II -
Abstract
With the popularization of e-commerce, the research about Online Consumer
Reviews has experienced a process from easy to difficult. In recent years, due to
the improvement of text mining technology, more and more scholars have turned
their attention to text reviews. However, there are contradictory conclusions in the
current study of the impact of text reviews on sales, which is likely due to
unreasonable text classification. Text classification depends on accurate sentiment
analysis methods, but the current Chinese sentiment analysis method needs further
improvement.
This article uses the big data mining tool Python to capture 12,278,861 pieces
of data from 109 tablets. Basing on the theory of optimal arousal theory, dilution
effect, bandwagon effect, persuasive effect and awareness effect, and according
to 839 consumers online review browsing path questionnaire survey results, this
paper uses the dynamic Panel model to regress the data. In order to improve the
accuracy of text classification, this paper improves the text mining technology
from three aspects, the selection of word segmentation tools, the expansion of the
thesaurus, and the classification of text comments. Through the choice of word
segmentation tools and the expansion of word banks, the accuracy of text
assignments is improved. Through the using of two-dimensional text classification
methods and according to consumer online shopping practices, the value range of
neutral text reviews is expanded, and the science of text classification is improved.
Online text reviews have been divided into positive text reviews, mixed positive
text reviews, mixed neutral text reviews, indifferent neutral text reviews, mixed
negative text reviews, and negative text reviews. Among them, positive text
reviews include positive text reviews and mixed positive text reviews. The results
show that positive text reviews have a “U”shape relationship with product sales,
mixed positive text reviews have a positive impact on product sales. Neutral text
reviews include mixed neutral text reviews and indifferent neutral text reviews.
The regression results show that mixed neutral text reviews have a positive impact
on product sales, while indifferent neutral text reviews have a negative impact on
product sales. The negative text reviews include mixed negative reviews and
negative text reviews. The results show that mixed negative text reviews have
positive impacts on product sales while negative text reviews have negative
impacts on product sales. The moderating effect of the number of reviews on the
relationship between text reviews and product sales. The results show that the
哈尔滨工业大学管理学硕士学位论文
- III -
moderating effect of the number of reviews appears to be an increase in product
sales. Among them, the number of reviews play a positive moderating effect on
the relationship between positive text reviews, mixed neutral text reviews, mixed
negative text reviews and product sales, and a negative moderating effect in the
relationship between indifferent neutral text reviews and product sales. The
research results have certain reference value for explaining the phenomenon of
conflicting research conclusions in the current online text reviews research field.
At the same time, it has a certain guiding role for the e-commerce platform to
maintain online reviews and promote product sales.
Keywords：text reviews, text sentiment analysis, two-dimensional text
classification, emotional dictionary
哈尔滨工业大学管理学硕士学位论文
- IV -
目录
摘要 ... I
ABSTRACT ...... II
第1章 绪论 ..1
1.1 研究背景与问题提出 .......... 1
1.1.1 研究背景 .......1
1.1.2 问题提出 .......2
1.2 国内外研究现状述评 .......... 3
1.2.1 评论数量对产品销量的影响研究现状 .........3
1.2.2 文本评论对产品销量的影响研究现状 .........3
1.2.3 在线评论情感分析研究现状 6
1.2.4 文献评述 .......8
1.3 研究目的与意义 ..... 9
1.3.1 研究目的 .......9
1.3.2 研究意义 ..... 10
1.4 研究内容与研究方法 ........ 11
1.4.1 研究内容 ..... 11
1.4.2 研究方法 ..... 12
1.5 研究创新点 ......... 13
第2章 理论框架与假设发展 . 15
2.1 理论框架 . 15
2.1.1 最佳唤醒理论 ......... 15
2.1.2 稀释效应 ..... 15
2.1.3 从众效应 ..... 15
2.1.4 知晓效应与说服效应 .......... 16
2.2 研究假设 . 16
2.2.1 评论数量对产品销量的直接作用 ... 16
2.2.2 文本评论对产品销量的作用 .......... 17
2.2.3 评论数量对文本评论与产品销量关系的调节作用19
2.3 研究框架 . 19
2.4 本章小结 . 20。