In this paper, we model the knowledge graph by constructing the new mechanism of tax collection and administration under the empowerment of IoT technology and give the knowledge subgraph and operation of the new mechanism of tax collection and administration. For the knowledge element extraction of the new mechanism of tax collection and administration, a dual-attention mechanism knowledge extraction model is constructed based on TransH coding and sentence feature attention model. Experiments on knowledge element extraction of the new mechanism of tax collection and administration are conducted for the model constructed in this paper. The results show that the model in this paper improves by 12.32%, 14.95% and 13.38% in accuracy, recall and F1 value, respectively, compared with the BiLSTM-CRF model. The model employed in this paper is effective in extracting the knowledge elements of the new tax collection and administration mechanism constructed based on the knowledge graph.