Detecting Fake Reviews Using N-gram Model and Chi-Square

Online product reviews are increasingly important in consumer purchasing decisions. They tend to refer to product or service reviews online before purchasing. Therefore, conducting more in-depth research in detecting fake reviews is essential. This research utilizes the n-gram model, especially bigrams and trigrams. The results of these two n-gram models were selected using the chi-square method. Next, the selected feature results are applied in three machine learning algorithms: Support Vector Machine, Random Forest, and Logistic Regression. SVM and Logistic Regression algorithms are suitable for use in binary classification, while Random Forest has the advantage of dealing with variations in the language and writing style of fake reviews. The best results in this research were obtained using the Support Vector Machine algorithm combined with trigrams and the chi-square method, with a value of K=1000. These results achieved an accuracy level of 92.19%. This achievement far exceeds previous research, which only achieved an accuracy of 88.75%. These results show that the approach of using word context and careful feature selection in detecting fake reviews improved the previous accuracy by 3.49%, even with the same algorithm.