Learning to Represent Review with Tensor Decomposition for Spam Detection

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 866–875, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Learning to Represent Review with Tensor Decomposition for Spam Detection Xuepeng Wang1,2, Kang Liu1, Shizhu He1 and Jun Zhao1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China {xpwang, kliu, shizhu.he, jzhao}@nlpr.ia.ac.cn Abstract Review spam detection is a key task in opin- ion mining. To accomplish this type of de- tection, previous work has focused mainly on effectively representing fake and non-fake re- views with discriminative features, which are discovered or elaborately designed by expert- s or developers. This paper proposes a nov- el review spam detection method that learn- s the representation of reviews automatically instead of heavily relying on experts’ knowl- edge in a data-driven manner. More specifi- cally, according to 11 relations (generated au- tomatically from two basic patterns) between reviewers and products, we employ tensor de- composition to learn the embeddings of the re- viewers and products in a vector space. We collect relations between any two entities (re- viewers and products), which results in much useful and global information. We concate- nate the review text, the embeddings of the re- viewer and the reviewed product as the rep- resentation of a review. Based on such rep- resentations, the classifier could identify the opinion spam more precisely. Experimental results on an open Yelp dataset show that our method could effectively enhance the spam detection accuracy compared with the state- of-the-art methods. 1 Introduction With the development of E-commerce, more and more customers share their experiences about prod- ucts and services by posting reviews on the we- b. These reviews could heavily guide the purchas- ing behaviors of customers. The products which receive more positive reviews tend to attract more consumers and result in more profits. Studies on Yelp.com have shown that an extra half-star rating could cause a restaurant to sell out 19% more prod- ucts (Anderson and Magruder, 2012), and a one- star increase leads to a 5-9% profit increase (Luca, 2011). Therefore, more and more sellers and manu- facturers have begun to place emphasis on analyzing reviews. However, the question remains: is every online review trustful? It has been reported that up to 25% of the reviews on Yelp.com could be fraudu- lent1. Due to the great profit or reputation, impostors or spammers energetically post fake reviews on the web to promote or defame targeted products (Jindal and Liu, 2008). Such fake reviews could mislead consumers and damage the online review websites’ reputations. Therefore, it is necessary and urgent to detect fake reviews (review spam). To accomplish this goal, much work has been conducted. They commonly regard this task as a classification task and most efforts are devoted to exploring useful features for representing target re- views. Li et al. (2013) and Kim et al. (2015) rep- resent reviews with linguistic features; Lim …

Content truncated to 500 words