In today’s data-driven era, deriving meaningful insights from documents poses a considerable challenge. The task of Document-level Relation Extraction (DocRE) seeks to discern relationships among multiple entities within documents, mirroring real-world scenarios and introducing more complexity than sentence-level relation extraction (SLRE). DocRE encounters issues such as numerous widely dispersed entities, an imbalance between positive and negative instances, and a long-tail distribution of relation types. We introduce a model for document-level relation extraction that utilizes boundary distance loss and long-tail relation enhancement to tackle these challenges. This approach adeptly captures intricate interactions among entities, enhancing multi-relation extraction performance. Our model achieved improvements of 3.22% and 3.15% in F1 and Ign-F1 scores on the RE-DocRED dataset, respectively. Moreover, it significantly surpasses recent strong baseline models on the JacRED dataset. The experimental results confirm the model’s effectiveness and generalizability.