A Study on Identification of Important Features for Efficient Detection of Fake Reviews
Ting-You Lin, Basabi Chakraborty, Chun‐Cheng Peng
Abstract
With the increasing advent of e-commerce, more and more people are engaged in online shopping. People rely on online reviews for making their purchasing decision as these reviews can provide a lot of useful information about the goods or services. These online reviews also help firms in understanding customer sentiment and behavior. However, fake or manipulated reviews are also posted to promote or demote the quality of the products or services which mislead the consumers and guide them to make wrong decision. Identification of fake review is difficult and their detection is currently an important issue. It is important to find out the characteristic feature which can be used for accurate detection of fake news. In this work, a framework for fake review detection has been proposed in which several types of features associated with the review, sentiment features, topic features and readability features are studied. The effectiveness of the features and their different combinations for discrimination of fake review from real review has been explored using different supervised classifiers. The simulation experiments are done with Amazon and Yelp benchmark review data set. It is found that the readability features are the most effective features. Some combination of readibility features and topic features can produce better performance with lesser number of features than only sentiment analysis for discriminating fake review from real one.