A Semi-Supervised Machine Learning-Based Detection of Deceptive Opinions in Digital Social Networks

Nidhi A. Patel, Nirali R. Nanavati

Deceptive Review, Machine Learning, Online Social Networks, Semi-supervised Learning, Spam Review, User-Generated Content

Online reviews play an important part in guiding consumers decisions in todays e-commerce and social networking environment. As the influence of Online Social Networks (OSNs) grows, so does the prevalence of deceptive or fraudulent reviews crafted to mislead customers, enhance product reputation, or undermine competitors. These fabricated opinions distort genuine user experiences and significantly impact sales and brand credibility. This work aims to accurately distinguish between factual and deceptive reviews using Positive and Unlabeled (PU) machine learning techniques within a semi-supervised framework. By integrating linguistic features and representation methods such as word2vec, trigram, bigram and unigram models, the proposed approach effectively identifies fraudulent content even with limited labelled data. Experimental results demonstrate improved accuracy and performance compared to traditional PU-based approach, highlighting the potential of advanced machine learning strategies for mitigating review spam and enhancing the reliability of online platforms.

https://dx.doi.org/10.31873/IJEAS.12.11.01