Bot Detection Using Multi-Input Deep Neural Network Model in Social Media

Recently, social media has been viewed as a critical tool for businesses to engage with other users, customers, and future consumers to build popularity, solicit ideas and opinions from users, or influence another user. This massive amount of data quickly becomes a valuable asset for businesses and organizations as a powerful tool for gaining insights and making critical decisions. Unfortunately, certain disinformation operations have been orchestrated by using bots, which are social media accounts operated by computer scripts that attempt to pass as actual human users to sway public opinion and disseminate misleading information. Bot identification is difficult since many bots deliberately strive to avoid detection. There are several approaches to distinguishing a bot from a legitimate account. This paper uses a multi-input deep neural network model with word embedding vectors to represent textual data for content-based bot detection, using fewer data and getting greater accuracy. Our proposed model, which includes two parallel convolutional neural network channels and fully connected neural networks to combine them, outperforms other newly proposed models in bot detection. We reached 93.25% accuracy in distinguishing between bot and human Twitter accounts using this novel technique.