Abstract: The nodes of the internet of things (IoT) have become the source of internal attacks due to their complex and unsupervised environment, which are easily be captured or damaged. In response to this situation, a dynamic comprehensive trust evaluation method based on the behavior of IoT nodes is proposed. This method calculated the direct evaluation value of the node based on the multiple behavior characteristics of the node during interaction, and the confidence and recommendation trust of nodes were proposed as the reference of weight coefficients. On this basis, the recommended evaluation value of the node was obtained through the common neighbor node to shorten the time for JP2discovering malicious nodes, and the time decay factor and historical evaluation window were used to ensure the timeliness JPof trust. The experimental results show that the malicious nodes can be identified timely through the abnormal behavior changes of nodes, which reduces the energy consumption of the network, and has good adaptability and resistance.