摘要
为了识别商品垃圾评论,基于垃评论员发表的多为垃圾评论这一基本思想,提出一种基于评论员评论行为来判定其是否为垃圾评论员的方法。分析定义了垃圾评论员常见的三类评论行为,分别是针对同类商品发表垃圾评论,针对同品牌商品发表垃圾评论和针对同一卖家商品发表垃圾评论;在对这三类评论行为建模的同时提出一种依据重复性过高或过低打分的评论数量来计算评论员垃圾指数(spam score)的方法。实验数据为在当当网摄影摄像商品区发表过评论的评论员的所有评论信息。实验结果通过人工评判和计算NDCG(normalize discounted cumulative gain)值的方法来检验,实验结果准确有效。
To detect product review spam on reviewers＇ behaviors, a method based on the idea that review spammers always issue product review spams is presented. Three characteristic behaviors of review spammers are identified and modeled, including targeting at product type, targeting at product brand and targeting at product seller. Meanwhile, scoring methods is proposed for the three behaviors to measure spam score of each reviewer based on his or her repeated overhigh or overlow rating. In experiments, the reviews are come from camera product reviewers of DANGDANG website. Manually evaluating and NDGG （nor malize discounted cumulative gain） calculating are adopted to judge our experimental results. The results show that our method is accuracy and effective on detecting review spammers.
出处
《计算机工程与设计》
CSCD
北大核心
2012年第11期4314-4319,共6页
Computer Engineering and Design
基金
北京林业大学新进教师科研启动基金项目(BLX2w8019)
中央高校基本科研业务费专项基金项目(YX2011-30)
