A Hybrid Deep Learning Model to Accurately Detect Anomalies in Online Social Media

Abstract

Online social media (OSM) generates a massive amount of data about human behavior based on their interactions. People express their opinions, comments and share information about variety of topics of their daily life through OSM. The majority of the comments are divided into three categories: Positive, negative, and natural. Regarding the negative comments, the OSM platform facilitates abnormal actions, such as unsolicited messages, misinformation, rumors, the dissemination of fake news and propaganda, as well as the dissemination of malicious links. Therefore, one of the most significant data analytic activities for identifying normal and deviant individuals on social networks is abnormality detection. This paper proposes a hybrid model based on three famous deep learning approaches to discover the behavioral abnormalities, and negative comments in the OSM platforms. The selected benchmark for our research is the airline sentiment in a Twitter dataset. In the proposed method, the dataset is fed to the LSTM network; next, the output of LSTM feeds the CNN network. CNN combines features and generates various feature maps. In the next step, we reduce and select the most important features. Finally, the selected features are given to ANN to classify the data.  The proposed method (LSTM + CNN + ANN) is compared with various classical machine-learning (ML) techniques. The experimental results show that the proposed method enhances the accuracy and precision on average by %8.6 and %8.4, respectively in compared to the classical ML techniques.