Lattice-based Integration of Knowledge Graph Information for Chinese NER
Xiaoxue Wang, Housheng Su · 2024
Knowledge Graph (KG) is a structured knowledge representation that encompasses abundant entity and relationship information, offering rich contextual insights for named entity recognition(NER). This study proposes the LIGNER model, which utilizes the Lattice-LSTM to integrate KG information into the process of NER for Chinese text. Lattice-LSTM is a neural network structure based on deep learning, which offers a flexible framework for organizing and representing entities. The LIGNER model integrates external information based on the Lattice framework, which helps to integrate structured knowledge from KG and improves the accuracy of NER. To address the challenge of low accuracy in long entity recognition, a new approach incorporating part-of-speech and position information is introduced. Experimental results on Chinese NER datasets demonstrate that LIGNER achieves advanced performance and efficiency compared with competitive methods.