Negative Deceptive Opinion Spam

Proceedings of NAACL-HLT 2013, pages 497–501, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics Negative Deceptive Opinion Spam Myle Ott Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14853 {myleott,cardie}@cs.cornell.edu Jeffrey T. Hancock Department of Communication Cornell University Ithaca, NY 14853 jeff.hancock@cornell.edu Abstract The rising influence of user-generated online reviews (Cone, 2011) has led to growing in- centive for businesses to solicit and manufac- ture DECEPTIVE OPINION SPAM—fictitious reviews that have been deliberately written to sound authentic and deceive the reader. Re- cently, Ott et al. (2011) have introduced an opinion spam dataset containing gold standard deceptive positive hotel reviews. However, the complementary problem of negative deceptive opinion spam, intended to slander competitive offerings, remains largely unstudied. Follow- ing an approach similar to Ott et al. (2011), in this work we create and study the first dataset of deceptive opinion spam with negative sen- timent reviews. Based on this dataset, we find that standard n-gram text categorization tech- niques can detect negative deceptive opinion spam with performance far surpassing that of human judges. Finally, in conjunction with the aforementioned positive review dataset, we consider the possible interactions between sentiment and deception, and present initial results that encourage further exploration of this relationship. 1 Introduction Consumer’s purchase decisions are increasingly in- fluenced by user-generated online reviews of prod- ucts and services (Cone, 2011). Accordingly, there is a growing incentive for businesses to so- licit and manufacture DECEPTIVE OPINION SPAM— fictitious reviews that have been deliberately writ- ten to sound authentic and deceive the reader (Ott et al., 2011). For example, Ott et al. (2012) has esti- mated that between 1% and 6% of positive hotel re- views appear to be deceptive, suggesting that some hotels may be posting fake positive reviews in order to hype their own offerings. In this work we distinguish between two kinds of deceptive opinion spam, depending on the sentiment expressed in the review. In particular, reviews in- tended to promote or hype an offering, and which therefore express a positive sentiment towards the offering, are called positive deceptive opinion spam. In contrast, reviews intended to disparage or slander competitive offerings, and which therefore express a negative sentiment towards the offering, are called negative deceptive opinion spam. While previous related work (Ott et al., 2011; Ott et al., 2012) has explored characteristics of positive deceptive opin- ion spam, the complementary problem of negative deceptive opinion spam remains largely unstudied. Following the framework of Ott et al. (2011), we use Amazon’s Mechanical Turk service to produce the first publicly available1 dataset of negative de- ceptive opinion spam, containing 400 gold standard deceptive negative reviews of 20 popular Chicago hotels. To validate the credibility of our decep- tive reviews, we show that human deception detec- tion performance on the negative reviews is low, in agreement with decades of traditional deception de- tection research (Bond and DePaulo, 2006). We then show that standard n-gram text categorization tech- niques can be used to detect negative deceptive opin- …

Content truncated to 500 words