Abstract

The way in which people purchase online products or services has greatly improved because of the development of Internet technologies. Customer purchase behavior has also modified as more people write reviews about items procured. As previously mentioned, when buyers intend to acquire products from online e-commerce websites, they often go through the details of the products such as ratings and initial posted reviews, instead of selecting a product based purely on price. Furthermore, they also give preference to the products that have higher score ratings and a high positive feedback. An online product with a higher percentage of positive reviews has a better chance than one which has a higher percentage of negative reviews. Unfortunately, some persons take advantage of the importance and value of posting online reviews by attempting to generate fake reviews, either to increase the promotion of specific products or to deceive customers into purchasing low-quality products. Skillful paid persons called “fraudsters” or “fake reviewers” aim to write deceptive positive and/or negative reviews. However, the problem of fake reviews can financially influence online business companies and primarily might be used to mislead customers into making wrong decisions, purchasing undesired products. Recently, fake reviews detection has captivated the attention of e-commerce, media, and governments. In the present paper, the researchers focus on detecting and classifying the online products reviews into fake or truthful via rules-based classifiers. For that purpose, the proposed methodology contains various phases, namely collection of dataset acquired from the Amazon platform, preprocessing, feature extraction, labeling process performed on extracted features, and finally machine learning as well as deep learning-based models adopted for the task of classification. When comparing the results obtained from the experiments carried out in this research work, it has been observed that the Convolutional Neural Networks incorporated in Bidirectional Long Short-Term Memory (CNN-BiLSTM) technique provides better performance than Random Forest (RF) technique in terms of accuracy metric.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call