Semantic Information Mining and Fusion Method for Bot Detection

Twitter bot detection is a crucial yet challenging task. The existing bot detection methods have limited semantic information mining ability for a large number of tweets posted by social media users. To address this challenge, we propose SIMF, which stands for Semantic Information Mining and Fusion. SIMF leverages a pre-training scorer to rank a large number of a user's tweets, and filters them based on specific rules. Additionally, SIMF preprocesses tweets and integrates multiple information encodings to obtain user representation, which enhances its ability to capture a diverse array of fake bots. Our comprehensive experiments on the benchmark TwiBot-20 demonstrate that SIMF outperforms other competitive algorithms.