Core Viewpoint - The article discusses the SIAMD framework for proactive bot detection based on structural information principles, emphasizing its effectiveness in combating misinformation and enhancing the authenticity of online interactions [3][4]. Group 1: Framework Overview - SIAMD consists of four main stages: social network analysis, network structure evolution, network content evolution, and bot detection optimization [6][12]. - The framework organizes user accounts and social messages into a unified heterogeneous structure, quantifying uncertainty through structural entropy [3][7]. Group 2: Social Network Analysis - Historical interactions between user accounts and social messages are extracted to construct a heterogeneous network, which is then used to distinguish between bot and human accounts [13][16]. - The network captures various types of interactions, including posting, retweeting, mentioning, replying, and following [16][17]. Group 3: Network Structure Evolution - The evolution of network structure involves analyzing user account influence and behavior correlation, modeling future behaviors of multiple bot accounts [8][32]. - Two behavior objectives for bot accounts are defined: minimizing detection probability and maximizing message propagation [8][34]. Group 4: Network Content Evolution - The framework generates relevant message content using large language models based on the constructed prompts from bot behaviors and interactions [39][40]. - The process integrates user metadata, historical content, and social structure to produce contextually appropriate messages [40][45]. Group 5: Detection Optimization - The bot detector is fine-tuned on the updated heterogeneous network to maximize the prediction probability of identifying bot accounts [10][36]. - Each iteration of optimization enhances the proactive detection capabilities of the model [10][36]. Group 6: Experimental Results - SIAMD was evaluated against several state-of-the-art baseline models using well-known bot datasets, demonstrating superior performance in accuracy, precision, recall, and F1 score [43][44]. - The framework consistently outperformed baseline models, achieving an accuracy of 98.6% on the Cresci-15 dataset and 90.7% on the TwiBot-20 dataset [44][43]. Group 7: Robustness and Interpretability - SIAMD's robustness was tested against attacks using large language models, showing minimal performance degradation compared to baseline methods [49][50]. - The framework's interpretability was validated through visualizations of bot behaviors and interactions within a community, highlighting the collaborative relationships among bot accounts [51][54].