<p>Document-level relationship extraction (DocRE) aims to identify all relationships between pairs of entities in a document, with the primary challenge being the aggregation of contextual information for effective relationship extraction. In this study, we propose a dynamic parsing and relationship inference model for entity features that incorporates gating units and pooling strategies to dynamically update hidden states. The approach fuses multiple mentions of entity features to construct a composite entity embedding that reflects the context, thereby reducing the loss of semantic information. Moreover, we combine self-attention and cross-attention mechanisms to integrate global entity semantic information across multiple rounds and capture complex interactions between entity pairs to enhance the prediction of multiple indirect relationships. Experimental results on three mainstream datasets, namely, DocRED, Chemical–Disease Relations, and Gene–Disease Associations, demonstrate that our model outperforms the baseline model Adaptive Thresholding and Localized Context Pooling, particularly in extracting potential relationships across sentence–boundary entity pairs.</p>