CDSCREM: A Cross-Domain Relation Extraction Model Integrating Syntax and Context

Cross-domain relation extraction, an emerging research task, evaluates the generalization ability of relation extraction models across different domains. However, existing studies lack a comprehensive and efficient model. To address this gap, this study proposes the Cross-Domain Syntactic and Contextual Relation Extraction Model (CDSCREM), which enhances cross-domain relation extraction by combining syntactic structure and contextual information. CDSCREM employs a domain aware syntactic graphical encoder, analyzing syntactic dependencies between entities using graph convolutional networks (GCN) to extract domain-specific syntactic information. Additionally, the model introduces an adaptive multidimensional attention mechanism to dynamically adjust attention to syntax and context, enhancing multidimensional data processing. The model also integrates a gated recurrent unit (GRU) for bi-directional context fusion, effectively combining forward and backward contextual information to enhance text-dependent capture. Results from the CrossRE 1.0 and 2.0 datasets demonstrate that CDSCREM significantly outperforms the baseline and other existing models across several test domains. Ablation experiments further confirm the key contribution of each component to the model's performance.