<p>This paper focuses on the task of document-level relation extraction (DocRE), which holds significant importance in the field of natural language processing. Traditional sentence-level relation extraction methods face limitations when dealing with relation facts distributed across multiple sentences. In contrast, document-level relation extraction aims to extract relations from documents containing multiple sentences and complex semantic scenarios. This paper first analyzes the challenges faced by document-level relation extraction, including the dispersion of entity mentions and the complexity of inter-sentence relations. To effectively address these challenges, this paper proposes an innovative model: DocChainGraphNet (DCGN). This model integrates mention nodes, sentence nodes, and document nodes, and captures the potential relation facts within the document through four different types of edges. Based on this heterogeneous graph, a chain-of-thought reasoning strategy is further introduced to mine multi-level logical associations and potential relations. Experiments on the DocRED and CDR datasets demonstrate that our DCGN model achieves strong performance, with F1 scores of 62.83% and 65.80% on the respective test sets. The experiments confirm the effectiveness of the proposed approach.</p>