Machine Learning Based Deception Detection System in Online Social Networks

Abstract

The rapid dissemination of Internet technologies makes it easier for people to live in terms of access to information. However, in addition to these positive aspects of the internet, negative effects cannot be ignored. The most important of these is to deceive people who have access to information whose reliability is controversial through social media. Deception, in general, aims to direct the thoughts of the people on a particular subject and create a social perception for a specific purpose. The detection of this phenomenon is becoming more and more important due to the enormous increase in the number of people using social networks. Although some researchers have recently proposed techniques for solving the problem of deception detection, there is a need to design and use high-performance systems in terms of different evaluation metrics. In this study, the problem of deception detection in online social networks is modeled as a classification problem and a methodology that detects misleading contents in social networks using text mining and machine learning algorithms is proposed. In this method, since the content is text-based, text mining processes are performed and unstructured data sets are converted to structured data sets. Then supervised machine learning algorithms are adapted and applied to the structured data sets. In this paper, real public data sets are used and Support Vector Machine, k-Nearest Neighbor (k-NN), Naive Bayes, Random Forest, Decision Trees, Gradient Boosted Trees, and Logistic Regression algorithms are compared in terms of many different metrics.