An Improved Bot Identification with Imbalanced Data using GG-XGBoost
Greeshma Lingam, B. Yasaswini, P.V.S.Lakshmi Jagadamba, Niharika Kolliboyana · 2022 2nd International Conference on Intelligent Technologies (CONIT) · 2022
In online social networks (OSNs), bots usually disseminate fraudulent information. In OSNs, when compared to the benign users, the bots are much less in number. However, the bot can create more negative impact in OSNs. One of the main research topics is to improve the accuracy of bot identification. Traditional bot identification approaches rely on supervised learning models. However, most supervised learning models are trained with a dataset which contains highly unbalanced distribution of data with benign users and bots. In this work, a generative adversarial network (GAN) with gated recurrent unit (GRU) is considered to overcome the unbalanced distribution of bots in OSNs. Further, we propose a novel algorithm namely GG-XGBoost by integrating GRU-GAN with XGBoost model. Experimentation have been performed on Twitter dataset to verify the efficiency of GG-XGBoost algorithm.