在线评论对商家销售业绩的影响：情感分析视角

The effect of awareness and persuasion that online reviews have on consumers result in influencing their purchasing decisions. However, many companies suffer from poor sale performance of improving products and adjusting price in time based on feedbacks, due to their incompetence of effectively identifying emotional information from massive reviews. Therefore, in a new perspective of sentiment analysis, this research follows the three stages of 'review posting, review adopting and review's impact', and combines methods of decision science, behavior analysis and econometrics to uncover the mechanism that how online reviews affect business performance. In the perspective of information technology, text mining methods will be used to unscramble online reviews in a fine-grained manner, by extracting product features and opinions, assessing sentiment information (polarity and intensity), and quantifying reviews in four dimensions as sentiment, quality, statistic and topic. In the perspective of behavioral study, empirical study on behaviors will be utilized to explore the reasons for customers posting reviews online with different sentiments, and analyze the influence that the emotional information has on review adoption of both customers and business. And in the perspective of economic analysis, econometric model of the relationships between online reviews and business performance will be established to explain the impact of product features and opinions on sales and price, hereby providing basis for product improvement and dynamic pricing. Besides that, experiments based on online reviews of both experience product and searchable product will be conducted to test and verify the effectiveness of this research. This research will enrich the studies on measuring the business value of online reviews theoretically, and help managing online reviews as well as implementing word-of-mouth marketing in business practically.

在线评论的知晓和说服效应影响消费者的购买意愿。然而商家没能有效识别海量评论的用户情感，无法根据用户反馈改进产品和调整价格，从而影响商家的销售业绩。 为此，基于情感分析的视角，沿着"评论发布->评论采纳->评论影响"主线，采用设计科学、行为分析和计量经济相结合的方法，勾画在线评论对销售业绩的影响路径。在技术层面，利用文本挖掘算法，提取产品特征及用户观点，判断情感类型，对评论的情感、质量、统计和主题四维属性进行量化，实现细粒度的评论解读；从行为角度，实证分析用户发表不同情感评论的动因，揭示情感类型对用户和商家采纳意愿的影响；从经济学视角，构建在线评论对商家业绩影响的计量模型，解释产品特征及评论属性对销量和价格的影响，为产品改进和动态定价提供依据；以体验型和搜索型产品的评论为实验对象，验证研究结果的有效性。 理论上，丰富在线评论价值发现的研究体系。实践上，为商家管理在线评论，实施口碑营销提供指导

网络口碑是影响消费者购买行为的重要因素，在线评论中的各种评价信息会改变用户对产品质量的感知，进而影响购买意愿。但是，在线评论对商家销售业绩的影响机理仍缺少系统性研究。为此，以在线评论为研究对象，基于情感分析的视角，综合行为科学、信息技术、计量分析三种范式，沿着“评论发布->评论采纳->评论影响”主线，勾画在线评论对销售业绩的影响路径，在此基础上，围绕以下内容展开研究。.（1）采用Python 语言编写多线程爬虫程序，结合深度优先和广度优先的搜索算法，对电商平台的商品信息、在线评论以及用户数据进行采集和预处理，并经人工标注，形成实验语料。.（2）从行为科学角度，基于调查问卷的实证方法，发现用户通过点评网站发表不同情感评论的动因，同时揭示用户通过点评网站获取评论并改进购买决策的意愿。.（3）从技术层面，考虑评论文本的信息特征度量和情感倾向的混合性，量化并抽取评论内容特征，采用GBDT模型评估特征集合分类效果，结合贪婪式特征选择算法识别有效内容特征，分析其对评论质量检测的影响。.（4）从技术层面，采用机器学习和本体建模方法，从评论中提取产品特征及用户观点，判断情感类型，并进行跨领域的鲁棒性检测，实现细粒度的评论解读；.（5）采用计量分析方法，构建在线评论对商家业绩的影响模型，解释产品特征及评论属性对销量的影响。同时，将计量模型拓展到股市波动、影视票房、上市公司业绩等预测方面，进一步验证模型有效性。.（6）采用计量经济方法，以消费者效用理论为基础，以网络零售两阶段销售为背景，建立与在线评论相关的产品定价模型，为商家动态定价提供依据。.（7）情感分析在产品评论商业价值挖掘中的应用：从在线评论中获取用户的比较性观点，进而形成产品间的竞争性关系，同时可以定位产品缺陷，为产品改进提供科学依据。.理论上，丰富在线评论价值发现的研究体系。实践上，为商家管理在线评论，实施口碑营销提供指导。