Chinese Nested Entity Recognition Benchmark in the Field of Naval Warfare

Named entity recognition (NER) is a fundamental task in natural language processing and a key technology for building knowledge graphs. However, the performance of NER is often limited by domain-specific characteristics. For unpopular Chinese entity recognition fields, such as naval warfare, NER typically performs poorly due to the lack of sufficient pre-trained models and labeled data. Furthermore, the naval warfare field includes numerous specialized terms and complex domain knowledge, with nested structures and physical relationships that are uncommon in conventional corpora, posing additional challenges for NER. To address these issues, we manually construct a high-quality naval warfare domain corpus. We also propose a new nested entity recognition framework, which fully leverages entity boundary information, enhances the learning of entity adjacency modeling information, and uses multi-scale convolution to obtain richer span feature representation. Experiments demonstrate the effectiveness of this method in four Chinese NER corpora.