Identifying fake Amazon reviews as learning from crowds

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 279–287, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Identifying fake Amazon reviews as learning from crowds Tommaso Fornaciari Ministero dell’Interno Dipartimento della Pubblica Sicurezza Segreteria del Dipartimento ComISSIT tommaso.fornaciari@interno.it Massimo Poesio University of Essex CSEE University of Trento CIMeC poesio@essex.ac.uk Abstract Customers who buy products such as books online often rely on other customers reviews more than on reviews found on specialist magazines. Unfortunately the confidence in such reviews is often mis- placed due to the explosion of so-called sock puppetry–authors writing glowing reviews of their own books. Identifying such deceptive reviews is not easy. The first contribution of our work is the cre- ation of a collection including a number of genuinely deceptive Amazon book re- views in collaboration with crime writer Jeremy Duns, who has devoted a great deal of effort in unmasking sock puppet- ing among his colleagues. But there can be no certainty concerning the other re- views in the collection: all we have is a number of cues, also developed in collab- oration with Duns, suggesting that a re- view may be genuine or deceptive. Thus this corpus is an example of a collection where it is not possible to acquire the actual label for all instances, and where clues of deception were treated as anno- tators who assign them heuristic labels. A number of approaches have been proposed for such cases; we adopt here the ‘learn- ing from crowds’ approach proposed by Raykar et al. (2010). Thanks to Duns’ cer- tainly fake reviews, the second contribu- tion of this work consists in the evaluation of the effectiveness of different methods of annotation, according to the performance of models trained to detect deceptive re- views. 1 Introduction Customer reviews of books, hotels and other prod- ucts are widely perceived as an important rea- son for the success of e-commerce sites such as amazon.com or tripadvisor.com. How- ever, customer confidence in such reviews is often misplaced, due to the growth of the so-called sock puppetry phenomenon: authors / hoteliers writing glowing reviews of their own works / hotels (and occasionally also negative reviews of the competi- tors).1 The prevalence of this phenomenon has been revealed by campaigners such as crime writer Jeremy Duns, who exposed a number of fellow au- thors involved in such practices.2 A number of sites have also emerged offering Amazon reviews to authors for a fee.3 Several automatic techniques for exposing such deceptive reviews have been proposed in recent years (Feng et al., 2012; Ott et al., 2001). But like all work on deceptive language (computational or otherwise) (Newman et al., 2003; Strapparava and Mihalcea, 2009), such works suffer from a seri- ous problem: the lack of a gold standard contain- ing ‘real life’ examples of deceptive uses of lan- guage. This is because it is very difficult to find definite proof that an Amazon review is either de- ceptive or genuine. Thus …

Content truncated to 500 words