基于知识采纳模型和多层感知机神经网络的评论有用性识别研究  

Identification of Usefulness for Online Reviews Based on Knowledge Adoption Model and Multilayer Perceptron Neural Network

作　　者：张婧[1] 周怡欣 胡涵 卞亦文[3] ZHANG Jing;ZHOU Yi-xin;HU Han;BIAN Yi-wen(School of Data Science and Artificial Intelligence,Dongbei University of Finance and Economics,Dalian 116024,China;School of Management Science and Engineering,Dongbei University of Finance and Economics,Dalian 116024,China;SILC Business School,Shanghai University,Shanghai 201899,China)

机构地区：[1]东北财经大学数据科学与人工智能学院,辽宁大连116024 [2]东北财经大学管理科学与工程学院,辽宁大连116024 [3]上海大学悉尼工商学院,上海201899

出　　处：《中国管理科学》2022年第4期264-274,共11页Chinese Journal of Management Science

基　　金：国家自然科学基金资助项目(71901053,72031004)。

摘　　要：移动互联网、社交媒体平台及电子商务的迅速发展,产生了大量的用户评论,其商业价值凸显,如何有效识别用户评论的有用性成为重要研究主题。本文提出基于知识采纳模型(Knowledge Adoption Model,KAM)理论和多层感知机(Multilayer Perceptron,MLP)神经网络的分类算法对评论文本进行有用性识别。该算法根据知识采纳模型理论从评论质量和评论来源可信度两方面进行评论有用性识别的特征提取:利用先验领域知识词典构造领域词占比、停用词占比等评论质量方面的特征,有效解决了特定领域评论存在的领域知识壁垒问题;根据评论作者的粉丝数、作者获赞数等信息构建评论来源可信度方面的特征。为了验证本文所提方法的识别效果,本文采用知乎论坛中医相关评论数据进行实验;实验结果表明,本文提出的方法能有效提高在线评论有用性的分类效果,提高了可解释性。With the rapid development of the mobile Internet,social media platforms and electronic commerce,a large amount of online reviews with user comments have been generated and show great business value.The recognition of usefulness of user comments provides an important guarantee for mining valuable information in comments.To this end,a classification algorithm is proposed based on the Knowledge Adoption Model(KAM)theory and Multilayer Perceptron(MLP)neural networks to identify the usefulness of online reviews in social media platforms.According to the knowledge adoption model theory,the algorithm extracts the featuresfor identifying usefulness of comments from two aspects:review quality and review source credibility.Specifically,to represent the review quality by constructing featuresusing the proportion of domain words and the proportion of stop words via a dictionary containing prior domain knowledge,which effectively alleviates the problem of domain knowledge barriers in cross-domain reviews;to represent the review source credibility by constructing the features based on the number of fans of the author and the number of likes the author has already received so far.In order to verify the recognition effect of the method proposed,traditional Chinese medicine related comments are crawled from Zhihu.com as the experimental data set.The experimental results show that the feature construction method,which integrates domain knowledge into text feature representation by calculating the proportion of domain words,can effectively alleviates the problems of knowledge barriers in specific fields;and the proposed method provides important information of review quality for the identification of the usefulness of online reviews,and thus has improved the identifying effect of usefulness of online comments and increasedthe interpretability to some extent.

关 键 词：在线评论 评论信息有用性 知识采纳模型 多层感知机神经网络 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]