管理评论

›› 2017, Vol. 29 ›› Issue (1): 144-154.

• 市场营销 • 上一篇    下一篇

网络口碑对产品销量的影响:基于细粒度的情感分析方法

孟园1, 王洪伟1, 王伟2   

  1. 1. 同济大学经济与管理学院, 上海 210000;
    2. 华侨大学工商管理学院, 泉州 362021
  • 收稿日期:2014-10-22 出版日期:2017-01-28 发布日期:2017-03-16
  • 通讯作者: 王洪伟(通讯作者),同济大学经济与管理学院教授,博士生导师,博士。
  • 作者简介:孟园,同济大学经济与管理学院博士研究生;王伟,华侨大学工商管理学院讲师,博士。
  • 基金资助:

    国家自然科学基金项目（71371144；71601119；71601082）；上海市哲学社会科学规划课题一般项目（2013BGL004）；江西省教育厅科学技术研究课题一般项目（GJJ150783）。

The Effect of Electronic Word-of-mouth on Sales through Fine-gained Sentiment Analysis

Meng Yuan1, Wang Hongwei1, Wang Wei2   

  1. 1. School of Economics and Management, Tongji University, Shanghai 210000;
    2. School of Business Administration, Huaqiao University, Quanzhou 362021
  • Received:2014-10-22 Online:2017-01-28 Published:2017-03-16

PDF (PC)

1728

摘要：

随着在线点评系统的发展，网络口碑成为消费者购买决策的重要参考依据，并对产品销量产生影响。护肤品销量受口碑效应的影响显著，为此以护肤品为例，基于细粒度情感分析技术，从网络口碑中提取针对产品特征项的消费者主客观情感，根据词汇频数设计细粒度情感各维度的权重指数，构建细粒度综合情感指数。然后，以综合情感指数和消费者评分为情感变量，结合ARMA模型对产品的销量预测进行实证分析。与基准模型对比，发现加入了情感变量的销量预测模型提高了对数据的拟合能力，细粒度情感指数有较高的预测精度。而消费者评分在某些节点上也具有一定的预测能力，但整体预测效果并不理想。研究也表明，以月度为观察周期构建的网络口碑综合情感指数具有较好的预测效果，综合情感指数滞后1期时能提供最好的预测效果，滞后1-4期时能为销量预测起作用。

关键词: 网络口碑, 销量预测, 情感指数, 细粒度情感分析

Abstract:

With the development of online reviews system, WOM has become an important reference for consumers' purchasing decision and affects product sales trend. Taking skin care products for example——in which WOM effect is most obvious, this paper makes use of fine-gained sentiment analysis technology to extract objective sentiment and subjective sentiment in the WOM. According to the "Term Frequency", the paper designs and tests the weight index of each dimension of fine-gained sentiments, and eventually constructs the fine-gained sentiment index. Then, fine-gained sentiment index and consumer rating are taken as input emotion variable in the ARMA model respectively for sales forecast empirical analysis. The comparative results show that the model which adds an emotion variable could improve the data fitting ability. The sentiment index has stronger predictive power and higher prediction precision. The model in which the consumer rating as an emotion variable has higher prediction ability on some point, but the overall predictive power isn't ideal. The results also show that sentiment index that is one period behind provides the best prediction effect. Sentiment index that is one to four periods behind could produce some effects on sales forecast.

Key words: electronic word-of-mouth, sales forecast, sentiment index, fine-gained sentiment analysis