Twitter Spam Detection Using Optimized Deep Model

Abstract

Social networking(SN) sites like twitter helps people to keep in touch with friends, extended family and stay connected with their professional networks for career opportunities. Social media spam attacks are a persistent challenge in today's world. It requires a multi-faceted approach like spam detection(SD) to ensure genuine users can engage without being bombarded by unwanted or harmful content. Spammer spreads multiple fake tweets that take links or hashtags on the website and online services. Businesses and academics have created spam-free SN platforms in several ways to maintain a positive user experience.Spammers create new spam every day and spread it across this platform. The existing spam protection methods failed to identify new spam and control the spam content. So, it is crucial to strengthen the performance of Twitter SD approaches. The ongoing problem of social media spam requires strong spam detection (SD) techniques to protect consumers from undesirable or dangerous content. This paper aims to rectify the deficiencies of current spam protection systems by proposing an enhanced deep learning (DL) technique for Twitter social dynamics (SD). The optimal deep learning model comprises a deep convolutional neural network (DCNN) that has been improved using the butterfly-optimizer algorithm (BOA). This study developed an optimised deep Learning(DL)methodto classify Twitter SD to strengthen the performance of a deep model. The optimised DL model is constructed with deep convolution neural network(DCNN), it is optimised using butterfly-optimizer algorithm(BOA). Then the optimized deep model trained data is used to classify the twitter spam using the support vector machine (SVM) classifiers. This optimized DCNN tained SVM classifier is named as DCNNOSVM model. TheDCNNOSVM approach detects the spammers by analysing tweet content and user meta-data such as account age and several followers. Theefficiency of the DCNNOSVM model is estimated, and the evaluation outcomesdemonstrated that the DCNNOSVM model attaineda maximum of 99.32% accuracy as classification accuracy.