一种情感分析与质量控制的异常评论识别方法

A spammer detection method based on sentiment analysis and quality control on comments

针对因数据量的增加以及异常评论策略的更新,以用户内容和行为为基础的传统微博异常评论识别方法效果不断下降的问题,提出一种基于情感分析和质量控制的微博异常评论识别方法.通过将预处理后的微博评论进行情感分析,将微博评论进行量化处理,在对微博评论进行质量控制的过程中,根据异常与正常用户在时域上对热点微博的评论分布差别检测可疑时间间隔,结合用户聚类分析,设计了异常评论识别模型.结果表明:该方法利用情感评分,对于评论文本进行较为准确的情感分类,然后通过调整边界值范围和时间阈值范围来限定异常检测等级,当边界值范围增大时,对于异常评论的检测范围扩大,容忍度下降,检测灵敏度高;当时间阈值扩大时,容忍度提高,检测灵敏度较低;适当的选择边界值和时间阈值,可以有效提高与正常评论行为相似的异常评论识别准确率.

To avoid the poor effect of spammer detection in traditional methods based on content and user behavior due to the increase of data and spammers’updated strategy,a spammer detection method based on sentiment analysis and quality control in microblog was proposed.In the method,the pre-processed comments in microblog were quantified by sentiment analysis.Then in the process of quality control of microblog commentary,the suspicious interval was detected according to different distribution between spammers and normal users for hotspot in varying time.Then a model for spammer recognition was established by cluster analysis.The experimental results showed that the method used the emotion score to make the emotion classification more accurately for the comment text,and then adjusted the boundary value range and time threshold range to limit the anomaly detection level.When the boundary value range increased,the anomaly detection range of the abnormal comment was increased,the tolerance was reduced and the detection sensitivity was improved.When the time threshold was expanded,the tolerance was improved and the detection sensitivity was reduced.Therefore,the appropriate choice of the boundary value and time threshold can effectively improve the accuracy of anomaly recognition which is similar to the normal commentary behavior.