Abstract
Fake profile detection is one of the critical problems in Online Social Networks (OSNs). So far, studies have mainly focused on profile-based, behavioral-based, network-based and content-based attributes. However, user sentiments have not been explored along this domain. In the present study, we proposed the fake profile detection model that incorporates sentimentbased attributes to differentiate real and fake OSN profiles. The study is grounded in the fact that the posts of real users reveal varied categories of emotions such as joy, sad, angry, fear, etc. based on their life experiences. On the contrary, fake users share posts to accomplish a specific purpose, and therefore, it is highly likely that their post content will contain similar types of emotions. The experiments are conducted on the posts of Facebook users. The detection model is trained on 12 emotion-based attributes including Plutchik's eight basic emotions, positivity and negativity. Furthermore, a noise removal technique is presented to remove the outliers from the dataset. Finally, several machine learning techniques including Support Vector Machine (SVM), Naive Bayes, JRip and Random Forest have been used to train the detection model.