Bot2Vec: A general approach of intra-community oriented representation learning for bot detection in different types of social networks

Abstract Recently, due to the rapid growth of o nline s ocial n etworks (OSNs) such as Facebook, Twitter, Weibo, etc. the number of machine accounts/social bots that mimic human users has increased. Along with the development of a rtificial i ntelligence (AI), social bots are designed to become smarter and more sophisticated in their efforts at replicating the normal behaviors of human accounts. Constructing reliable and effective bot detection mechanisms is this considered crucial to keep OSNs clean and safe for users. Despite the rapid development of social bot detection platforms, recent state-of-the-art systems still encounter challenges which are related to the model’s generalization (and whether it can be adaptable for multiple types of OSNs) as well as the great efforts needed for feature engineering. In this paper, we propose a novel approach of applying network representation learning (NRL) to bot/spammer detection, called Bot2Vec. Our proposed Bot2Vec model is designed to automatically preserve both local neighborhood relations and the intra-community structure of user nodes while learning the representation of given OSNs, without using any extra features based on the user’s profile. By applying the intra-community random walk strategy, Bot2Vec promises to achieve better user node embedding outputs than recent state-of-the-art network embedding baselines for bot detection tasks. Extensive experiments on two different types of real-word social networks (Twitter and Tagged) demonstrate the effectiveness of our proposed model. The source code for implementing the Bot2Vec model is available at: https://github.com/phamtheanhphu/bot2vec