Deceptive Review Detection using Supervised Machine Learning Detection

Abstract

As the digital landscape continues to thrive, online marketing has gained tremendous popularity. The easy accessibility of products and services has prompted customers to rely heavily on reviews to guide their purchasing decisions. Unfortunately, this reliance has opened the door for fraudulent practices, with unscrupulous individuals fabricating reviews to serve their own interests. These deceptive reviews not only mislead customers but also hinder their ability to make well-informed choices. Consequently, there is a pressing need to identify and eliminate such deceitful content.

In this dissertation, we present an innovative approach to detect deceptive reviews by harnessing the power of linguistic, POS, behavioural, and textual features. Our aim is to enhance the accuracy and dependability of online review systems, empowering consumers to navigate the digital marketplace with confidence.

Drawing upon a vast dataset of yelp reviews, we leverage cutting-edge machine learning techniques to evaluate our approach. By scrutinizing linguistic features such as nouns, adjectives, and verbs, we uncover the distinct linguistic patterns exhibited by deceptive reviews. POS features, encompassing prepositions, determiners, and connector words, enable us to capture the linguistic cues that betray fraudulent intent. Furthermore, our analysis of behavioural features, including sentiment scores, review frequency, and rating deviations, provides invaluable insights into user behaviour, uncovering anomalous patterns indicative of deceptive practices. To complete the picture, we examine textual features such as review length, word count, and the presence of emojis, shedding light on the structural and stylistic aspects of deceptive reviews.

The results of our comprehensive experiments are highly encouraging, revealing an impressive accuracy of 89.20% and an F1 score of 94.28%. These outcomes firmly establish the effectiveness of our multi-feature approach in identifying deceptive reviews and elevating the quality of consumer information. By fusing diverse features, we offer a holistic view of review data, empowering customers to make informed decisions in the dynamic realm of online commerce.