Exploring groups of opinion spam using sentiment analysis guided by nominated topics.

[Display omitted] • This is the first study using platform-offered aspects for spam detection. • This is also the first study to address Chinese reviews for aspect-oriented fraud labelling. • Cross-platform reviews are integrated for spam group clustering. • We achieve high performance in spam clustering. Currently, it is common to see untruthful opinions (also known as review spam, fraud or shilling attack) that resemble each other explicitly or implicitly across multiple business-to-customer websites or opinion sharing communities. Unfortunately, these fake recommendations can be fabricated by individual spammers or results of a manipulation campaign. Considering its severe harmfulness in influencing a product's reputation, grouped spam is more urgent to detect than individual fraud. Most state-of-the-art techniques of labeling grouped spam, e.g., Frequent Itemset Mining (FIM) or Latent Dirichlet Allocation (LDA), are completely unsupervised and incapable of making good use of officially recommended topics, such as appearance , speed and standby are three suggested aspects along a cell phone product in JD.com. In this paper, we introduce a novel approach based on aspect-oriented sentiment mining that can identify spam groups supported by nominated topics. Experiments show that our method is effective and outperforms several state-of-the-art solutions with statistical significance on two metrics, content duplication and burstiness of time. [ABSTRACT FROM AUTHOR]