在线商品评论有用性主题分析及预测研究

Online product reviews helpfulness prediction based on topic analysis

随着电子商务的飞速发展,电商平台上的在线商品评论成为消费者在线购物时做出购买决策的重要参考,同时也是平台商家获取在线消费者真实关切的重要信息来源.然而,海量的良莠不齐的在线商品评论使得消费者和商家很难从中获取有价值的高质量信息.一方面,本文在经典的主题分析LDA模型的基础之上提出了一种基于评论有用性的主题分析模型,即Help-LDA模型.相比与假定每条评论具有同等重要程度的LDA模型,Help-LDA模型根据评论有用性对不用评论赋予不同的权重,进而从有用性较高的评论中抽取出对于消费者更有用的决策信息.另一方面,本文基于Help-LDA模型提出了新的评论文本表示方法,并结合SVM方法进行评论有用性预测.通过收集大众点评网站在线评论进行的实验表明,Help-LDA模型能够从电商评论中高质量抽取在线消费者对于商家商品和服务的真实关切.并且基于Help-LDA模型的评论文本表示结合SVM方法能够显著提升在线评论有用性预测性能.

With the prosperity of e-commerce,online reviews have become an important information source for both the online consumers and vendors in their decision making.However,with the marvelous reviews on the e-commerce platform,it is hard for consumers and vendors to acquire valuable information to support their decision making.This paper proposes a novel topic model called help-LDA by extending the classic LDA model with considering the helpfulness of online reviews.On the one hand,the proposed help-LDA model can extract helpful topics from online reviews.On the other hand,the proposed help-LDA model can be used for online review representation with goal of predicting the helpfulness of online reviews.With the real data collected from Dianping.com,we conduct extensive experiments to compare the proposed help-LDA model and the baseline models in topic modeling and helpfulness prediction of online reviews.The experimental results demonstrate the superiority of the proposed help-LDA model over the baseline models.