搜索型商品在线评论有用性影响因素研究

发布时间：2018-04-09 12:46

本文选题：搜索型商品　切入点：在线评论有用性　出处：《江苏大学》2017年硕士论文

【摘要】：随着电子商务的迅猛发展,消费者使用网络购物已渐渐成为常态化,由于网络购物的信息不对称,商品在线评论成为人们获取商品和服务信息的重要来源,同时也是消费者分享商品使用体验的平台。由于在线评论存在大量性、多样性和复杂性等特点,客观上要求及时、正确挖掘出有效评论信息。因此,探究在线评论有用性的影响因素,并在此基础上建立评论有用性识别模型,为今后建立更加灵活有效的评论评价系统具有重要的指导意义。本文以抓取到的亚马逊购物网站6种搜索型商品共计1042条在线评论为研究对象,以信息采纳模型、有用信息的特征以及消费者购买行为为理论基础,在已有文献的基础上,从评论内容属性和评论者属性两个方面构建评论有用性影响因素模型。然后以评论长度、评论内容完整性、评论及时性、评论情感强度和评论来源的可信度为自变量,其中评论内容完整性又细分为服务信息、物流信息和商品信息,以评论有用性为因变量,通过建立Tobit回归模型来探究评论有用性的影响因素,并对验证通过的因素进行相关性分析。最后根据评论有用性的影响因素建立评价指标体系,选取朴素贝叶斯算法、支持向量机算法和C4.5决策树算法分别建立分类模型,将评论分为“有用”和“无用”两类评论,并根据分类模型分类的准确率和效率,选取最优分类模型。研究结果表明,评论字数越多、评论中提及商家或平台的服务信息、评论中提及的商品属性越多、主观性表达越丰富、评论星级评分越低、评论者排名越靠前,则消费者评论的感知有用性越高,而评论中是否提及物流信息和评论发表的天数对评论有用性的影响不显著;评论长度与评论内容的完整性、评论主观性呈正相关关系,与评论情感强度无关,评论者排名与评论长度、评论中的商品信息呈负相关关系,与评论情感强度无关;支持向量机算法的分类精确度最高,综合准确率达到74.28%,F1测度值为0.736,分类过程耗时0.28秒,因此,选择支持向量机算法作为搜索型商品在线评论有用性的分类模型。
[Abstract]:With the rapid development of electronic commerce, consumers' online shopping has gradually become a norm. Because of the information asymmetry of online shopping, online commodity review has become an important source for people to obtain information about goods and services.At the same time, it is also a platform for consumers to share the experience of using goods.Due to the large quantity, diversity and complexity of online comments, it is necessary to mine the effective comment information correctly and timely.Therefore, it is of great significance to explore the factors influencing the usefulness of online reviews and to establish a model for the identification of comment usefulness, which will be helpful to the establishment of a more flexible and effective comment evaluation system in the future.In this paper, a total of 1042 online reviews of 6 kinds of search items on Amazon shopping website are taken as the research object. Based on the information adoption model, the characteristics of useful information and the purchasing behavior of consumers, and on the basis of the existing literature, this paper makes a research on the information adoption model of Amazon shopping website.The influence factors model of comment usefulness is constructed from two aspects: comment content attribute and reviewer attribute.Then the length of the comment, the integrity of the content of the comment, the timeliness of the comment, the emotional intensity of the comment and the credibility of the source of the comment are taken as independent variables, in which the integrity of the comment content is subdivided into service information, logistics information and commodity information.Based on the dependent variable of the usefulness of comments, the Tobit regression model is established to explore the influencing factors of the usefulness of comments, and the correlation analysis of the factors verified by the model is carried out.Finally, the evaluation index system is established according to the influential factors of comment usefulness, and the naive Bayesian algorithm, support vector machine algorithm and C4.5 decision tree algorithm are selected to establish classification models respectively, and the comments are divided into two categories: "useful" and "useless".According to the accuracy and efficiency of classification model, the optimal classification model is selected.The results showed that the more the number of words in the comments, the more the service information of the merchant or platform was mentioned in the comments, the more attributes of goods were mentioned in the comments, the more subjective expression was, the lower the rating of the comments, the higher the ranking of the reviewers.The perceived usefulness of the consumer comment is higher, but whether or not the logistics information and the number of days published in the comment have no significant influence on the comment usefulness, the length of the comment is positively correlated with the integrity of the content of the comment, and the subjectivity of the comment is positively correlated.It has nothing to do with the emotional intensity of the comment, the rank of the reviewer is negatively related to the length of the comment, the information of the product in the comment is negatively correlated with the emotional intensity of the comment, and the support vector machine algorithm has the highest classification accuracy.The synthetic accuracy rate of 74.28 / F _ 1 is 0.736, and the classification process takes 0.28 seconds. Therefore, support vector machine (SVM) algorithm is selected as the classification model for the usefulness of online reviews of search products.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.36;F713.55

【参考文献】