In this paper, we select the structure of Cell Simultaneous Recurrent Neural Network (CSRN), which is good at dealing with two-dimensional structural data processing, as a prediction method for multilevel information data, and explain its network structure and operation principle. It also describes the structure of four important elements, namely, unit state, forgetting threshold, input threshold and output threshold, in the special long and short-term memory network of recurrent neural networks. The multilevel information is transformed into multimodal information, and the data information of different modalities is analyzed using the projection tracing method. Combine with recursive neural network algorithm to construct a multilevel information extraction model. Comparing the extraction performance of similar modeling algorithms on multimodal information data, the designed multilevel information extraction model performs the best in all indicators on data set-1. The F1 value of 86.34%, the precision rate of 88.84%, recall rate of 88.22% and F1 value of 88.52% in Marco-F1 values show excellent multilevel information extraction performance.
