Spotting Fake Reviews via Collective Positive-Unlabeled Learning

Abstract

Abstract-Online reviews have become an increasingly important resource for decision making and product designing. But reviews systems are often targeted by opinion spamming. Although fake review detection has been studied by researchers for years using supervised learning, ground truth of large scale datasets is still unavailable and most of existing approaches of supervised learning are based on pseudo fake reviews rather than real fake reviews. Working with Dianping 1 , the largest Chinese review hosting site, we present the first reported work on fake review detection in Chinese with filtered reviews from Dianping's fake review detection system. Dianping's algorithm has a very high precision, but the recall is hard to know. This means that all fake reviews detected by the system are almost certainly fake but the remaining reviews (unknown set) may not be all genuine. Since the unknown set may contain many fake reviews, it is more appropriate to treat it as an unlabeled set. This calls for the model of learning from positive and unlabeled examples (PU learning). By leveraging the intricate dependencies among reviews, users and IP addresses, we first propose a collective classification algorithm called Multi-typed Heterogeneous Collective Classification (MHCC) and then extend it to Collective Positive and Unlabeled learning (CPU). Our experiments are conducted on real-life reviews of 500 restaurants in Shanghai, China. Results show that our proposed models can markedly improve the F1 scores of strong baselines in both PU and non-PU learning settings. Since our models only use language independent features, they can be easily generalized to other languages.

Fake Comments Detection Method Based on Integrated Features

“…Li proposed a collective classification algorithm named MHCC (Multi-typed Heterogeneous Collective Classification) [6]. The advantage of MHCC is that it can transfer the user suspicious probability to their comments, which is the intermediate result MHCC algorithm.…”

“… CPU (Collective PU-Learning): This method is an improvement of PU-LEA NB [6]. The difference between these two methods is that CPU regards the IP information and user's own information as the classification features which is still the static characteristics.…”

“…Li etc. made a further improvement of the method and put forward a kind of collective classification algorithms named MHCC (Multi-typed Heterogeneous Collective Classification) by adding users and IP address information, then extended it with the PULearning, obtaining a better prediction result [6][7].…”

Spam Detection Framework Using Sentimenatl Analysis

“…[1] Then again, a lot of writing has been distributed on the systems used to recognize spam and spammers and in addition distinctive sort of investigation on this theme. These systems can be classi fied into various classifications; some utilizing phonetic examples in content, which are for the most part in view of bigram, and unigram [2], others depend on behavioral examples that depend on highlights removed from designs in clients' conduct which are generally metadata based, and even a few procedures utilizing diagrams and chart based calculations and classifiers [3]. The way that anyone with a character winds up endorsed to comment a review, makes an alluring open entryway for spammers to make fake studies arranged particularly to mislead customer's inclination.…”

Impact of reviewer social interaction on online consumer review fraud detection

“…Typically, no user profile is available but additional information for products and for reviews are available depending on site. The problem of opinion fraud detection has been addressed by focusing on the review of text information [10,31], and by behavioral approaches that utilize the behavior of fake users [18,19,24]. The problem has often been approached as a supervised classification problem with two classes, fraud and not-fraud.…”