Event extraction is a key technology in natural language processing and has strong application prospects for extracting knowledge from unstructured data. The current event extraction technique is mainly based on sentence-based event extraction, which has the disadvantages of incomplete coverage of extracted events and ambiguity in event classification. In this paper, we propose the ERNIE-BiGRU-CRF model for chapter-level event extraction, which encodes the semantic enhancement of paragraph text by ERNIE pretrained language model, inputs a bidirectional gated neural network for feature extraction, and finally obtains the annotated sequence by CRF layer. In this paper, we perform event extraction on the Baidu financial domain documen-level event extraction dataset using the sequence-labeled trigger extraction model and the sequence-labeled event element extraction model, and the results show that the final model's F1 value is 5.45 percentage points higher than the baseline model.