Identification of fake reviews using semantic and behavioral features

摘要

In recent years, online reviews have been playing an important role in making purchase decisions. This is because, these reviews can provide customers with large amounts of useful information about the goods or service. However, to promote factitiously or lower the quality of the products or services, spammers may forge and produce fake reviews. Due to such behavior of the spammers, customers would be misleaded and make wrong decisions. Thus detecting fake (spam) reviews is a significant problem. In this paper, we propose two types of features and apply supervised machine learning algorithms for performing classification on Yelp's real-life data. In terms of features used, there are two new semantic feature sets: readability features and topic features. Our results show that our proposed new features are more effective than n-gram features in detecting spam reviews. To improve classification on the real Yelp review data, we use a set of behavioral features about reviewers and their reviews for learning, which dramatically improves the classification result on real-life opinion spam data. For further improvement, we also ensure the number of reviewers instead of reviews is balanced.