A Method for Cultural Relics Named Entity Recognition Based on Enhanced Lexical Features

Yao Li, He Yan, Yunxiang Yang, Xu Wang

Abstract: Named entity recognition (NER) of cultural relic data faces challenges in identifying entity boundaries and types due to the unique nature of word formation for these entities. To address this issue, we presents a method for named entity recognition of cultural relics based on enhanced lexical features. Specifically, this study introduces a multi-level temporal convolutional module at the encoding layer to encode the text, capturing lexical features of different granularities. In addition, we utilize the attention mechanism to combine lexical features with character embedding features of BERT in order to obtain feature information that enhances the lexical features.Finally, we employs a conditional random field (CRF) to determine the entity types. Comparative experiments conducted on a self-built cultural relic dataset, CRner, and two public datasets, Weibo and Resume, demonstrate that our model has a significant performance advantage over the baseline model and also has a degree of generalizability.

External IDs:dblp:conf/ijcnn/LiYYW24