The intelligent processing of Party history literature faces significant challenges due to the extensive use of pseudonyms,  alternative designations, and complex implicit relationships. This study proposes a multi-strategy semantic understanding and dynamic knowledge graph reasoning-based method for ambiguous reference resolution to address three major challenges in this field: the semantic gap, temporal evolution, and sparse evidence. The method constructs a domain-specific lexicon covering over ten thousand entities and a pseudonym-real name mapping database to incorporate prior knowledge. A domain dictionary-guided negative sample sampling strategy is employed to fine-tune pre-trained language models, enhancing their semantic perception of specific expressions. Finally, a time-constrained graph neural network reasoning algorithm is applied on a self-built temporal knowledge graph to mine implicit relationships and perform consistency verification. Experimental results demonstrate that the proposed method achieves an overall F1 score of 80.6% on authoritative evaluation metrics, significantly outperforming existing baseline models,and effectively uncovers deep historical correlations. The research outcomes have been integrated into a visual prototype system, providing a reliable intelligent tool for Party history research.

ambiguous reference resolution , party history literature , temporal knowledge graph , graph neural network , pre-trained language model