A Machine Learning Model for Spam Reviews and Spammer Community Detection

Abstract

People’s choices to purchase a product are influenced by its internet ratings and recommendations. Spammers manipulate product sales by creating fraudulent ratings on online social media platforms. The majority of current research on online review has concentrated on supervised learning algorithms, which require labelled data.. This is an insufficient requirement for online review. In this piece, we will be concentrating on identifying any misleading text reviews that we come across. The goal of this study is to discover spam comments and spammer groups. Various spam detection strategies have been proposed in the literature, including Review-Linguistic (RL) based features, User-Behavioral (UB) based features, and Review-Behavioural (RB) based features, but none of them include a simultaneous detection of these characteristics and relative importance of the features while also defining the communication between spam users. The suggested work establishes a diverse network of users and feedback nodes, and then applies the spam detection methodology to the issue of the communication environment. A feature weighting approach is presented to determine the relative value of features. Our solution uses an attention mechanism to discover the spamming hints hidden within the material and determines the relevance of each word in the text by computing its weight. We used the CNN algorithm to classify the reviews and compared the results with the usual Naive Bayes and Support Vector Machine algorithms.