A Meta-data Based Feature Selection Mechanism to Identify Suspicious Reviews

Deceptive reviews on internet platforms are a harsh reality in today's world. Businesses and products are praised or vilified through reviews. In an online or web society, users get help from reviews before making a decision and in a similar way, web reviews are also very helpful for organizations to keep them updated as per customer needs. Many efforts through various algorithms have already been made to detect deceptive reviews. It has been observed that instead of focusing only on algorithms to find optimal predictions, it is also necessary to pay attention towards the selection of effective features. Datasets used for machine learning are loaded with a large number of additional features which are usually not required. This paper is presenting the significance of the meta-data-based features in order to predict the fake and genuine reviews. In the domain of finding suspicious reviews, various methodologies used for the selection of the optimal features are presented in the paper with the help of the amazon dataset. The target of this approach is to ensure the use of noise-free, relevant, non-redundant, optimal features of the dataset which will result in better predictions.