Identifying Manipulated Offerings on Review Portals

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1933–1942, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Identifying Manipulated Offerings on Review Portals Jiwei Li School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA jiweil@cs.cmu.edu Myle Ott Claire Cardie Department of Computer Science Cornell University Ithaca, NY 14853, USA myleott,cardie@cs.cornell.edu Abstract Recent work has developed supervised meth- ods for detecting deceptive opinion spam— fake reviews written to sound authentic and deliberately mislead readers. And whereas past work has focused on identifying individ- ual fake reviews, this paper aims to identify offerings (e.g., hotels) that contain fake re- views. We introduce a semi-supervised man- ifold ranking algorithm for this task, which relies on a small set of labeled individual re- views for training. Then, in the absence of gold standard labels (at an offering level), we introduce a novel evaluation procedure that ranks artificial instances of real offer- ings, where each artificial offering contains a known number of injected deceptive reviews. Experiments on a novel dataset of hotel re- views show that the proposed method outper- forms state-of-art learning baselines. 1 Introduction Consumers increasingly rely on user-generated online reviews when making purchase deci- sions (Cone, 2011; Ipsos, 2012). Unfortunately, the ease of posting content to the Web, potentially anonymously, combined with the public’s trust and growing reliance on opinions and other information found online, create opportunities and incentives for unscrupulous businesses to post deceptive opinion spam—fraudulent or fictitious reviews that are deliberately written to sound authentic, in order to deceive the reader (Ott et al, 2011). Unlike other kinds of spam, such as Web (Martinez-Romo and Araujo, 2009; Castillo et al, 2006) and e-mail spam (Chirita et al, 2005), recent work has found that deceptive opinion spam is neither easily ignored nor easily identified by human readers (Ott et al, 2011). Accordingly, there is growing interest in developing automatic, usually learning-based, methods to help users identify deceptive opinion spam (see Section 2). Even in fully-supervised settings, however, automatic methods are imperfect at identifying individual deceptive reviews, and erroneously labeling genuine reviews as deceptive may frustrate and alienate honest reviewers. An alternative approach, not yet considered in previous work, is to instead identify those prod- uct or service offerings where fake reviews appear with high probability. For example, a hotel manager may post fake positive reviews to promote their own hotel, or fake negative reviews to demote a com- petitor’s hotel. In both cases, rather than identify- ing these deceptive reviews individually, it may be preferable to identify the manipulated offering (i.e., the hotel) so that review portal operators, such as TripAdvisor or Yelp, can further investigate the sit- uation without alienating users.1 Accordingly, this paper addresses the novel task of identifying manipulated offerings, which we frame as a ranking problem, where the goal is to rank offerings by the proportion of their reviews that are believed to be deceptive. We propose a novel three- layer graph model, based on manifold ranking …

Content truncated to 500 words