Opinion Fraud Detection in Online Reviews by Network Effects


User-generated online reviews can play a significant role in the success of retail products, hotels, restaurants, etc. However, review systems are often targeted by opinion spammers who seek to distort the perceived quality of a product by creating fraudulent reviews. We propose a fast and effective framework, FRAUDEAGLE, for spotting fraudsters and fake reviews in online review datasets. Our method has several advantages: (1) it exploits the network effect among reviewers and products, unlike the vast majority of existing methods that focus on review text or behavioral analysis, (2) it consists of two complementary steps; scoring users and reviews for fraud detection, and grouping for visualization and sensemaking, (3) it operates in a completely unsupervised fashion requiring no labeled data, while still incorporating side information if available, and (4) it is scalable to large datasets as its run time grows linearly with network size. We demonstrate the effectiveness of our framework on synthetic and real datasets; where FRAUDEAGLE successfully reveals fraud-bots in a large online app review database. Introduction The Web has greatly enhanced the way people perform certain activities (e.g. shopping), find information, and interact with others. Today many people read/write reviews on merchant sites, blogs, forums, and social media before/after they purchase products or services. Examples include restaurant reviews on Yelp, product reviews on Amazon, hotel reviews on TripAdvisor, and many others. Such user-generated content contains rich information about user experiences and opinions, which allow future potential customers to make better decisions about spending their money, and also help merchants improve their products, services, and marketing. Since online reviews can directly influence customer purchase decisions, they are crucial to the success of businesses. While positive reviews with high ratings can yield financial gains, negative reviews can damage reputation and cause monetary loss. This effect is magnified as the information spreads through the Web (Hitlin 2003; Mendoza, Poblete, and Castillo 2010). As a result, online review systems are attractive targets for opinion fraud. Opinion fraud involves reviewers (often paid) writing bogus reviews (Kost May 2012; Copyright c © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Streitfeld August 2011). These spam reviews come in two flavors: defaming-spam which untruthfully vilifies, or hypespam that deceitfully promotes the target product. The opinion fraud detection problem is to spot the fake reviews in online sites, given all the reviews on the site, and for each review, its text, its author, the product it was written for, timestamp of posting, and its star-rating. Typically no user profile information is available (or is self-declared and cannot be trusted), while more side information for products (e.g. price, brand), and for reviews (e.g. number of (helpful) feedbacks) could be available depending on the site. Detecting opinion fraud, as defined above, is a non-trivial and challenging problem. Fake reviews are often written by experienced professionals who are paid to write high quality, believable reviews. As a result, it is difficult for an average potential customer to differentiate bogus reviews from truthful ones, just by looking at individual reviews text(Ott et al. 2011). As such, manual labeling of reviews is hard and ground truth information is often unavailable, which makes training supervised models less attractive for this problem. Summary of previous work. Previous attempts at solving the problem use several heuristics, such as duplicated reviews (Jindal and Liu 2008), or acquire bogus reviews from non-experts (Ott et al. 2011), to generate pseudo-ground truth, or a reference dataset. This data is then used for learning classification models together with carefully engineered features. One downside of such techniques is that they do not generalize: one needs to collect new data and train a new model for review data from a different domain, e.g., hotel vs. restaurant reviews. Moreover feature selection becomes a tedious sub-problem, as datasets from different domains might exhibit different characteristics. Other feature-based proposals include (Lim et al. 2010; Mukherjee, Liu, and Glance 2012). A large body of work on fraud detection relies on review text information (Jindal and Liu 2008; Ott et al. 2011; Feng, Banerjee, and Choi 2012) or behavioral evidence (Lim et al. 2010; Xie et al. 2012; Feng et al. 2012), and ignore the connectivity structure of review data. On the other hand, the network of reviewers and products contains rich information that implicitly represents correlations among these entities. The review network is also invaluable for detecting teams of fraudsters that operate collaboratively on targeted products. Our contributions. In this work we propose an unsuperProceedings of the Seventh International AAAI Conference on Weblogs and Social Media

Extracted Key Phrases

12 Figures and Tables

Citations per Year

104 Citations

Semantic Scholar estimates that this publication has 104 citations based on the available data.

See our FAQ for additional information.

Cite this paper

@inproceedings{Akoglu2013OpinionFD, title={Opinion Fraud Detection in Online Reviews by Network Effects}, author={Leman Akoglu and Rishi Chandy and Christos Faloutsos}, booktitle={ICWSM}, year={2013} }