TCAE-DL-RGCN Based Detection of Twitter Robots
Hebing Du, Chunling Wu, Pan He, Hongyang Li · 2024
Twitter bot detection is a challenging task that has received widespread attention in the field of social security. Detection of Twitter bots using graph neural networks has been widely used, graph neural networks are prone to over-smoothing phenomenon affecting the model accuracy when extracting features from the data so the number of network layers is usually small and there is insufficient feature extraction when extracting features from the data. To address the above challenges, our research proposes the TCAE-DL-RGCN model, which fuses the RGCN features extracted by the attention mechanism of controlled attention temperature with the RGCN extracted features to prevent over-smoothing and to improve the feature expression capability and model accuracy. Our extensive experiments on the public Twitter robot detection dataset Twibot-20 show that TCAE-DL-RGCN achieves optimal performance.