Multilayer Perceptron Optimization Approaches for Detecting Spam on Social Media Based on Recursive Feature Elimination

Abstract

Social Media Platforms are becoming increasingly popular around the world. Various social media platforms allow people to discuss their personal activities, opinions, and beliefs. Simultaneously, social spam is becoming increasingly prevalent and in a variety of styles across prominent social media platforms. As a result, spam identification seems to have become a significant and well-known issue. Users can use social media sites to stay virtually connected with their friends. Because of the growing popularity of social networking sites, users can now collect a vast amount of sensitive data about their connections. Twitter is the fastest-growing platform among a slew of others. Because of its popularity, several spammers have taken advantage of it by sending massive amounts of spam to legitimate users’ accounts. In this work, a model is proposed to detect the Spam users’ profiles on Twitter network. This work is built upon the user behavior-based and content-based features like Retweets, Mentions, Replies, Hashtags, and URLs. In this work, the three Multilayer Perceptron optimization approaches are used along with the Recursive Feature Elimination namely Stochastic Gradient Descent (SGD), Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS), and Adam. Weka has been used for data pre-processing, and implemented the Recursive Feature Elimination with the three optimization approaches mentioned using sklearn in Python. For the proposed model performance evaluation, the performance metrics such as Accuracy, Precision, TP Rate, FP Rate and F-Measure are used.