融合情感极性和逻辑回归的虚假评论检测方法

发布时间：2019-08-24 09:55

【摘要】：在线购物评论为消费者比较商品的质量和其他一些购买特性提供了有用信息,然而却有大量的虚假评论者受利益驱使撰写虚假或者不公正的评论来迷惑消费者。先前的研究一般都是使用文本相似度和评分模式来探测虚假评论,这些算法可以检测特定类型的攻击者,在现实场景中许多虚假评论者刻意模仿正常用户对商品进行评论,因此先前的算法对检测这类攻击效果不佳。本文通过分析评论文本的感情极性,抽取不同的特征并使用逻辑回归模型来检测虚假评论;首先,借用自然语言处理的相关技术来分析评论文本的情感极性,判断每个用户的情感偏离大众情感的程度,如果偏离越大则说明其是虚假评论者的概率就越大;然后再选取其他几个重要特征结合逻辑回归模型进行虚假检测;通过实验对比,表明了该方法取得了较好的效果。
[Abstract]:Online shopping reviews provide useful information for consumers to compare the quality of goods and other purchasing characteristics, but a large number of false commentators are driven by interests to write false or unfair comments to confuse consumers. Previous studies have generally used text similarity and scoring patterns to detect false comments. These algorithms can detect specific types of attackers. In real situations, many false commentators deliberately mimic normal users to comment on goods, so the previous algorithm is not effective in detecting such attacks. In this paper, the emotional polarity of the comment text is analyzed, different features are extracted and the logical regression model is used to detect the false comment. Firstly, the emotional polarity of the comment text is analyzed by using the related technology of natural language processing to judge the degree of emotional deviation of each user, and the greater the deviation, the greater the probability that it is a false commentator. Then several other important features are selected to carry out false detection combined with logical regression model, and the experimental results show that the method has achieved good results.
【作者单位】：山东师范大学信息科学与工程学院;山东省分布式计算软件新技术重点实验室;
【基金】：国家自然科学基金项目(61373149,61472233) 山东省科技计划项目(2012GGX10118,2014GGX101026) 山东省教育科学规划项目(ZK1437B010)
【分类号】：TP391.1;F724.6

【参考文献】