基于多维度文本特征的电商平台评论有用性研究

伴随电子商务的发展，基于Web 2.0的在线评论体系不断完善，但海量评论降低了有效信息的获取效率，引发信息过载问题。如何迅速识别对消费者有用的评论、缓解信息过载问题，引起了业内与学界的广泛关注。利用大众点评网近5万条餐厅在线评论数据，综合考虑评论的多维度文本特征与情感表达倾向，采用Tobit、负二项回归方法对评论有用性影响因素开展实证研究。在此基础上，提出餐厅在线评论有用性分类阈值并使用支持向量机算法检验阈值的分类性能。研究结果表明：在线评论内容的菜品品质维度属性数量与平均长度、商家服务维度属性数量与平均长度以及正向情感倾向均显著正向影响在线评论有用性，且阈值为2的模型分类性能最佳。因此，电子商务网站可以进一步优化在线评论体系，提高消费者信息获取效率，促进在线评论生态持续向好发展。

Research on Help Fulness of E-commerce Platform Reviews Based on Multi-dimensional Text Features

The flourishing development of e-commerce in China has gradually improved the online review system based on Web 2.0. However, the massive number of online reviews reduces the efficiency of obtaining effective information, leading to the problem of information overload. How to identify helpful reviews for consumers and alleviate the problem of information overload has aroused widespread concern in the industry and academia. Using nearly 50 000 online restaurant reviews from Dianping.com, this paper comprehensively considered the multi-dimensional text features and emotional tendency of reviews, and conducted an empirical study on the influencing factors of review helpfulness based on Tobit regression and negative binomial regression. On this basis, this paper proposed a classification threshold for the helpfulness of Chinese restaurant online reviews, and utilized support vector machine algorithm to examine performance of the threshold. The results show that the number and average length of dish quality dimension attributes of reviews, the number and average length of restaurant service dimension attributes of reviews and positive sentiment tendencies all significantly affect the helpfulness of online reviews. In addition, the model with a threshold of 2 gives the best classification result among all threshold models. Therefore, e-commerce websites can further optimize online review system to improve the efficiency of consumer information acquisition, and promote the continuous development of Chinese online review ecosystem.