A Lexicon Enhanced Chinese Long Named Entity Recognition Using Word-Aware Attention

In recent years, due to the rapid growth of space science and utilization research of China Manned Space Engineering, a considerable amount of technical documents and web data have been produced. Named entity recognition (NER) plays a vital role in extracting valuable information from these resources. However, the presence of numerous long named entities, which consist of complex, specialized terms and diverse phrase combinations, poses significant challenges for existing methods in identifying them correctly. To resolve this issue, we introduced two modules, SkipWord-Lattice and Word-Aware Attention, to improve the widely used lexical enhancement model in Chinese NER. SkipWord-Lattice reduces the overlap, redundancy, and confusion of word tokens, while Word-Aware Attention enhances the semantic interaction between character tokens and word tokens, improving the model’s comprehension of word tokens. Collectively, these modules significantly increase the model’s ability to identify long named entities. Moreover, using relevant corpus data publicly available in the field of space science and utilization of China Manned Space Engineering, we built a Chinese NER dataset, named SSUIE-NER, comprising rich entity types and a substantial number of long named entities. Experimental results indicate that, our method significantly improves the recognition of long named entities, outperforming other state-of-the-art (SOTA) approaches on SSUIE-NER and other three benchmarks in Chinese NER.