Towards Detecting the Fake Profile in Social Media using Hybrid Model of LSTM and Extreme Learning Machine

The last decade has seen an explosion in the use of online social networks (OSNs), with many people relying on them for the majority of their social contacts. They use OSNs to maintain relationships, disseminate information, plan events, and even do business online. Attackers and imposters looking to steal personal information, distribute fake news, or engage in other forms of bad behavior have taken notice of the explosive expansion of OSNs and the wealth of information about its members. However, academics have begun exploring effective approaches to identify suspicious or fraudulent accounts by analyzing account data and applying classification algorithms. In addition, results from stand-alone categorization algorithms aren’t always accurate, and some of the exploited account attributes actually have a detrimental impact on the final conclusions or have no impact at all. The solution provided makes use of preprocessing methods like tokenization and the removal of stop words. Chi-squared features are used to select and extract features before an LS TM-ELM model is trained. The proposed method outperforms the previous two models significantly.