【作者】 张灿；

【导师】 栾贻会；

【作者基本信息】 山东大学 ， 应用统计（专业学位）， 2019， 硕士

【摘要】 随着大数据时代的到来,越来越多的人通过互联网分享自己的观点和想法,用户在线评论数量呈指数型爆发,评论的控制和利用成为当前网络平台面临的重要考验。一个有效的网络评论管理系统应当具备以下两方面的功能:帮助用户快速的从海量数据中得到有用信息和帮助平台合理有效的管理和利用用户评论。评论质量评估作为自然语言处理的一个分支,成为网络评论管理系统的重要组成部分。评论质量评估即寻找可衡量评论质量的指标,根据相应指标对评论质量进行量化,进而可以根据质量高低将评论进行过滤、排序等更多处理,识别出质量较高的评论,使得阅读评论的人能够在海量评论中快速获取有价值的信息。对非商业化图书交流平台进行评论质量评估,一方面,有助于识别出高质量评论,使读者更加快速高效地发现有价值的评论,协助其选择适合自己的、更优质的书籍。另一方面,能够改进图书门户网站的现有评论展示功能,改善网站的服务质量,提高用户体验度。本文面向非商业化图书平台的用户评论进行了质量评估研究。首先分析了非商业化图书平台的特点,结合中文表达方式的特殊性,构建了一套适用于该类型平台的WDC在线评论质量评价指标体系,然后以该指标为基础分析了使用支持向量机方法、逻辑回归方法进行分类的可行性。最终,以“豆瓣读书”网站上三种图书类型的评论数据进行了实证分析,分别利用支持向量机方法和逻辑回归方法建立了在线评论质量评价模型,从查准率、召回率、F值、准确率四个方面对模型的分类效果进行了分析,发现在这套评价体系下,支持向量机方法的分类效果比逻辑回归方法更显著。同时采用随机森林方法对各指标进行排序后得出结论:对于非商业化图书平台上的用户评论,评论的修饰词数对评论质量影响最大,其次是平均句长和字符数,而评分差异对评论质量的影响最小。本研究的创新之处在于:一、针对目前研究较少的非商业化平台构建了评论质量评价指标体系;二、在标注训练集评论质量时,采用了有用性投票和人工标注相结合的方式,且在标注时与以往文本长度越长,有用性越强的理论不同,适中数据才会被判别有用,这一改进丰富了有用性的定义。本文的研究成果丰富了非商业化平台在线评论质量评估的研究内容,为后续研究做出了一定铺垫。更多还原

【Abstract】 With the arrival of big data’s era.more and more people share their views and ideas through the Internet.Online comments of users are exponentially explosive.The control and utilization of comments has become an important test faced by current network platforms.A good network comment management system should have two functions,one is to help users get useful information from massive data quickly,the other is to help the platform manage and utilize user comments reasonably and effectively.As a branch of natural language processing,comment quality assessment has become an important part of network comment management system.The purpose of comment quality assessment is to quantify the quality of comments through some measurable indicators or to classify comments according to their quality and identify high-quality comments.On this basis,comments are filtered,sorted,so as to help readers obtain valuable comment information quickly.On the one hand,the review quality evaluation of non-commercial book platform can help readers quickly and efficiently find valuable reviews in the massive review information and help them select more suitable and high-quality books.On the other hand,it can improve the existing review display function of the book portal,improve the service quality of the website and improve the user experience.This paper studies the quality assessment of user reviews on non-commercial book platforms.Firstly,according to the characteristics of non-commercial book platforms and Chinese language,a set of online comment quality evaluation index system--WDC comment quality evaluation index system suitable for this type of platform is constructed.Then,based on this index,the feasibility of classification by SVM method and logistic regression method is analyzed.In the end,with"douban reading"site three types of book reviews data has carried on the empirical analysis,support vector machine(SVM)method and logistic regression were used respectively to establish the online reviews quality evaluation model.To analyze the classification effect,precision rate,recall rate,F value,and accuracy rate are used.In this evaluation system,we find the effect of SVM classification method is superior to logistic regression method.At the same time,the random forest method was adopted to sort the indicators and the conclusion was drawn:for the user comments on the non-commercial book platform.the number of modifiers in the comments had the largest impact on the comment quality,followed by the average sentence length and the number of characters,and the difference in the score had the least impact on the comment quality.The innovation of this study is to construct a comment quality evaluation system for non-commercial platforms with less research before.And when labeling the quality of the training set,the combination of useful voting and manual labeling is adopted.Furthermore,in the past,the longer the comment is,we will consider it more useful,but in this label the moderate data will be judged useful.This improvement enriches the definition of usefulness.The research results of this paper enrich the research content of the online evaluation quality evaluation of non-commercial platform,which has laid a foundation for the follow-up research.更多还原