Detecting Deceptive Reviews Using Text Features Only

The growth of online reviews in the healthcare sector makes it a great challenge to distinguish between honest and deceptive reviews. This underlines the urgent need for tools to detect misleading reviews in order to ensure credibility. In this study, we address the detection of deceptive reviews in the healthcare sector by utilizing a dataset from Amazon's health and personal care reviews. We applied a machine learning pipeline that includes data cleaning, feature extraction, and classification. We focused only on text features such as N-grams, VADER sentiment analysis, and part-of-speech tagging. Our contribution is to compare the accuracy of different classification models, including logistic regression, naive bayes, and LSTM, combined with different text features, and identify an effective method for detecting deceptive reviews. Our results show that Bigram and Trigram models with Naive Bayes classifiers outperform other methods, achieving the highest accuracy in identifying deceptive reviews.