Abstract:

In today's business and commerce landscape, online reviews wield significant influence. Consumers heavily rely on user reviews when making purchasing decisions for products online. Unfortunately, this reliance has led to the rise of opportunistic individuals and groups attempting to manipulate product reviews for their own benefit. To combat this issue, a research paper introduces various text mining models, both semi-supervised and supervised, that aim to detect fake online reviews. The study also compares the effectiveness of these techniques using a dataset known as "Gold Standard." The focus of this research work was on implementing unsupervised machine learning algorithms, such as the expectation maximization-based naive Bayes (EM-NB) and expectation maximization-based support vector machine (EM-SVM). Additionally, supervised machine learning algorithms like NB and SVM models were utilized. To extract features from the dataset, the researchers employed the term frequency-inverse document frequency (TF-IDF) method, which helps uncover relevant properties related to the reviews. The extracted features using TF-IDF were then used to train all the models. After conducting simulations, the results showed that the proposed supervised SVM model outperformed the conventional EM-NB, EM-SVM, and supervised NB models in terms of detecting fake online reviews. This outcome highlights the potential of supervised learning techniques in effectively identifying and addressing fraudulent reviews, thereby bolstering the credibility of online reviews and aiding consumers in making informed decisions.

Keywords: